jcmarchi - Tumblr blog

jcmarchi · 18 minutes

Text

5 Best AI Apps for Couples (April 2024)

New Post has been published on https://thedigitalinsider.com/5-best-ai-apps-for-couples-april-2024/

5 Best AI Apps for Couples (April 2024)

In the age of artificial intelligence, couples are discovering innovative ways to strengthen their relationships and foster deeper connections. From AI-powered dating apps that prioritize compatibility to virtual relationship coaches offering personalized guidance, technology is changing the way couples navigate the complexities of love and partnership.

In this blog post, we’ll explore the top AI apps designed to help couples enhance communication, build intimacy, and create lasting bonds.

Image: Flamme

Flamme is an innovative platform that goes beyond traditional dating apps by offering personalized advice and insights to help couples strengthen their bond and overcome challenges. With its advanced AI algorithms, Flamme analyzes user data to provide tailored recommendations that cater to each couple’s unique needs and preferences.

What sets Flamme apart is its comprehensive approach to relationship support. The app not only helps users navigate the early stages of dating but also offers guidance for maintaining long-term relationships. From communication tips to date night ideas, Flamme has it all covered. The AI-powered love guru acts as a virtual relationship coach, empowering couples to grow together and build a stronger connection.

Key Features of Flamme:

Personalized Relationship Roadmap: Flamme creates a customized plan for each couple based on their goals, personalities, and relationship history.

Emotion Analyzer: The app’s AI technology can detect emotional cues in conversations, helping users better understand their partner’s feelings and needs.

Date Night Generator: Flamme suggests creative and engaging date ideas tailored to each couple’s interests and preferences.

Relationship Health Monitor: The app tracks key indicators of relationship health and provides proactive advice to help couples address potential issues before they escalate.

Image: Maia

Maia takes a fresh approach to couples’ apps by focusing on the power of meaningful conversations. This innovative platform uses AI-driven prompts to encourage partners to engage in deep, thought-provoking discussions that foster emotional intimacy and understanding.

One of Maia’s standout features is its ability to learn and adapt to each couple’s unique dynamics. The more users interact with the app, the more personalized the prompts and suggestions become. This ensures that every couple receives a tailored experience that addresses their specific needs and challenges.

Key Features of Maia:

AI-Powered Conversation Starters: Maia generates thought-provoking questions and topics to help couples engage in meaningful discussions.

Relationship Insights: The app analyzes user interactions to provide valuable insights into communication patterns, emotional needs, and areas for growth.

Mood Tracker: Maia helps users track their emotional well-being and provides personalized suggestions for self-care and relationship maintenance.

Relationship Milestones: The app celebrates important milestones and achievements, fostering a sense of progress and shared accomplishment.

Ringi is an AI app that aims to simplify relationship maintenance for busy couples. With just five minutes of daily interaction, Ringi provides a convenient and effective way to nurture and strengthen your bond with your partner. The app’s intuitive interface and personalized features make it easy to integrate into your daily routine, ensuring that your relationship remains a top priority.

What makes Ringi unique is its focus on practicality and efficiency. The app understands that modern couples often struggle to find time for lengthy relationship exercises or therapy sessions. By condensing key relationship-building activities into bite-sized, five-minute interactions, Ringi ensures that even the busiest couples can invest in their partnership. The app’s AI-powered tools adapt to your specific needs, providing targeted advice and activities that maximize the impact of every interaction.

Key Features of Ringi:

Five-Minute Relationship Booster: Ringi offers a daily five-minute activity designed to strengthen your connection with your partner.

AI-Powered Advice: The app’s AI technology provides personalized guidance and insights based on your unique relationship dynamics.

Relationship Health Snapshot: Ringi offers a quick and easy way to assess the overall health of your relationship, identifying areas for improvement.

Progress Tracker: The app tracks your relationship growth over time, celebrating your successes and helping you stay motivated.

Multi-Language Support: Ringi is available in English and Japanese, with plans to expand to other languages in the future.

Image: Relish

Relish is a relationship coaching app that harnesses the power of AI to help couples strengthen their bond and overcome challenges. Designed by a team of experienced relationship experts, Relish combines AI-driven analysis with engaging quizzes, activities, and personalized guidance to provide couples with the tools they need to build a thriving partnership.

At the heart of Relish’s approach is its ability to assess a couple’s communication patterns and interactions using advanced AI technology. By analyzing this data, the app generates tailored advice and exercises that address each couple’s specific needs and challenges. From improving conflict resolution skills to increasing intimacy and maintaining a healthy relationship, Relish offers comprehensive support every step of the way.

Key Features of Relish:

Expert-Crafted Content: Relish’s quizzes, activities, and educational materials are designed by licensed relationship therapists and coaches, ensuring a solid foundation in proven relationship science.

Personalized Coaching: The app’s AI-powered coaching adapts to each couple’s unique dynamics, providing targeted guidance and support.

Accessible and Affordable: Relish makes professional relationship support more accessible and affordable compared to traditional in-person couples therapy.

Couples’ Engagement: Interactive features encourage couples to complete exercises and activities together, fostering a sense of teamwork and strengthening their bond.

Iris Dating AI is an innovative dating app that takes a fresh approach to helping couples find their perfect match. By prioritizing physical attraction and compatibility, Iris aims to create more meaningful connections between users. The app’s advanced AI algorithms analyze user preferences, physical attributes, and compatibility factors to suggest potential matches, ensuring a higher likelihood of mutual attraction and long-term success.

What sets Iris Dating AI apart is its unique “training” process for new users. Upon joining the app, users are shown images of potential matches and asked to indicate their level of interest. This data is then used by the AI to refine the user’s preferences and provide more accurate match recommendations. By emphasizing mutual attraction from the start, Iris reduces the chances of mismatches and dead-end conversations, saving users time and frustration.

Key Features of Iris Dating:

AI-Powered Matchmaking: Iris’s advanced algorithms analyze user data to suggest highly compatible matches based on physical attraction and shared interests.

Attraction-Based Training: New users undergo a “training” process to help the AI better understand their preferences and refine match recommendations.

Streamlined Interface: The app’s user-friendly design allows for quick and easy identification of mutual interest, facilitating more efficient connections.

Compatibility Assessments: Iris evaluates a range of factors to ensure matched couples have a strong foundation for a lasting relationship.

Success Stories: The app showcases real-life couples who found love through Iris, inspiring users and demonstrating the effectiveness of its approach.

The Power of AI in Strengthening Relationships

As technology continues to advance, AI-powered apps like the ones in this blog are changing the way couples approach relationship building and maintenance. By leveraging the power of artificial intelligence, these innovative tools provide personalized insights, tailored advice, and engaging experiences that help couples deepen their connection and navigate the complexities of modern relationships.

From facilitating meaningful conversations and providing expert guidance to streamlining the dating process and offering accessible coaching, these AI apps are empowering couples to invest in their partnerships and build stronger, more resilient bonds. As more couples embrace the potential of AI-driven relationship support, we can expect to see a new era of more fulfilling, successful, and long-lasting relationships.

0 notes

jcmarchi · 5 hours

Text

Nobody Likes a Know-It-All: Smaller LLMs are Gaining Momentum

New Post has been published on https://thedigitalinsider.com/nobody-likes-a-know-it-all-smaller-llms-are-gaining-momentum/

Nobody Likes a Know-It-All: Smaller LLMs are Gaining Momentum

Phi-3 and OpenELM, two major small model releases this week.

Created Using Ideogram

Next Week in The Sequence:

Edge 391: Our series about autonomous agents continues with the fascinating topic of function calling. We explore UCBerkeley’s research on LLMCompiler for function calling and we review the PhiData framework for building agents.

Edge 392: We dive into RAFT, UC Berkeley’s technique for improving RAG scenarios.

You can subscribed to The Sequence below:

TheSequence is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

📝 Editorial: Nobody Likes a Know-It-All: Smaller LLMs are Gaining Momentum

Last year, Microsoft coined the term ‘small language model’ (SLM) following the publication of the influential paper ‘Textbooks Are All You Need’, which introduced the initial Phi model. Since then, there has been a tremendous market uptake in this area, and SLMs are starting to make inroads as one of the next big things in generative AI.

The case for SLMs is pretty clear. Massively large foundation models are likely to dominate generalist use cases, but they remain incredibly expensive to run, plagued with hallucinations, security vulnerabilities, and reliability issues when applied in domain-specific scenarios. Add to that environments such as mobile or IoT, which are computation-constrained by definition. SLMs are likely to fill that gap in the market with hyper-specialized models that are more secure and affordable to execute. This week we had two major developments in the SLM space:

Microsoft released the Phi-3 family of models. Although not that small anymore at 3.8 billion parameters, Phi-3 continues to outperform much larger alternatives. The model also boasts an impressive 128k token window. Again, not that small, but small enough 😉

Apple open-sourced OpenELM, a family of LLMs optimized for mobile scenarios. Obviously, OpenELM has raised speculations about Apple’s ambitions to incorporate native LLM capabilities in the iPhone.

Large foundation models have commanded the narrative in generative AI and will continue to do so while the scaling laws hold. But SLMs are certainly going to capture an important segment of the market. After all, nobody likes a know-it-all ;)”

🔎 ML Research

Phi-3

Microsoft Research published the technical report of Phi-3, their famous small language model that excel at match and computer science task. The new models are not that small anymore with phi-3-mini at 3.8B parameters and phi-3-small and phi-3-medium at 7B and 14B parameters respective —> Read more.

The Instruction Hierarchy

OpenAI published a paper introducing the instruction hierarchy which defines the model behavior upon confronting conflicting instructions. The method has profound implications in LLM security scenarios such as preventing prompt injections, jailbreaks and other attacks —> Read more.

MAIA

Researchers from MIT published a paper introducing Multimodal Automated Interpretability Agent (MAIA), an AI agent that can design experiments to answer queries of other AI models. The method is an interesting approach to interpretability to prove generative AI models to undestand their behavior —> Read more.

LayerSkip

Meta AI Research published a paper introducing LayerSkip, a method for accelerated inference in LLMs. The method introduces modification in both the pretraining and inference process of LLMs as well as a novel decoding solution —> Read more.

Gecko

Google DeepMind published a paper introducing Gecko, a new benchmark for text to image models. Gecko is structured as a skill-based benchmark that can discriminate models across different human templates —> Read more.

🤖 Cool AI Tech Releases

OpenELM

Apple open sourced OpenELM, a family of small LLMs optimized to run on devices —> Read more.

Artic

Snowflake open sourced Artic, an MoE model specialized in enterprise workloads such as SQL, coding and RAG —> Read more.

Meditron

Researchers from EPFL’s School of Computer and Communication Sciences and Yale School of Medicine released Meditron, an open source family of models tailored to the medical field —> Read more.

Cohere Toolkit

Cohere released a new toolking to accelerate generative AI app development —> Read more.

Penzai

Google DeepMind open sourced Penzai, a research tookit for editing and visualizing neural networks and inject custom logic —> Read more.

🛠 Real World ML

Fixing Code Builds

Google discusses how they trained a model to predict and fix build fixes —> Read more.

Data Science Teams at Lyft

Lyft shared some of the best practices and processes followed for building its data science teams —> Read more.

📡AI Radar

Perplexity announced it has $63 million at over $1 billion valuation.

Elon Musk’s xAI is closing in on a $6 billion valuation.

Microsoft and Alphabet beat Wall Street expectations with strong earnings fueled by AI adoption.

NVIDIA is acquiring AI ifnrastructure startup Run:ai for a reported $700 million.

Cognition, the startup behind coding assistant Devin, raised a $175 million round at $2 billion valuation.

Salesforce announced released Einstein Copilot Actions to bring actionability to its AI platform.

Adobe introduced Firefly 3 with new image generation capabilities.

Higher than expected AI investments had a negative impact in Meta’s earnings report.

Augment emerged from stealth mode with a monster $227 million round.

AI-biotech company Xaira Therapeutics launched with $1 billion in funding.

AI sales platform Nooks raised $22 million.

Snorkel AI announced major generative AI updates to its Snorkel Flow platform.

Flex AI raised $30 million for a new AI compute platform.

The OpenAI Fund closed a $15 million tranche.

TheSequence is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

0 notes

jcmarchi · 13 hours

Text

Exploring the history of data-driven arguments in public life

New Post has been published on https://thedigitalinsider.com/exploring-the-history-of-data-driven-arguments-in-public-life/

Exploring the history of data-driven arguments in public life

Political debates today may not always be exceptionally rational, but they are often infused with numbers. If people are discussing the economy or health care or climate change, sooner or later they will invoke statistics.

It was not always thus. Our habit of using numbers to make political arguments has a history, and William Deringer is a leading historian of it. Indeed, in recent years Deringer, an associate professor in MIT’s Program in Science, Technology, and Society (STS), has carved out a distinctive niche through his scholarship showing how quantitative reasoning has become part of public life.

In his prize-winning 2018 book “Calculated Values” (Harvard University Press), Deringer identified a time in British public life from the 1680s to the 1720s as a key moment when the practice of making numerical arguments took hold — a trend deeply connected with the rise of parliamentary power and political parties. Crucially, freedom of the press also expanded, allowing greater scope for politicians and the public to have frank discussions about the world as it was, backed by empirical evidence.

Deringer’s second book project, in progress and under contract to Yale University Press, digs further into a concept from the first book — the idea of financial discounting. This is a calculation to estimate what money (or other things) in the future is worth today, to assign those future objects a “present value.” Some skilled mathematicians understood discounting in medieval times; its use expanded in the 1600s; today it is very common in finance and is the subject of debate in relation to climate change, as experts try to estimate ideal spending levels on climate matters.

“The book is about how this particular technique came to have the power to weigh in on profound social questions,” Deringer says. “It’s basically about compound interest, and it’s at the center of the most important global question we have to confront.”

Numbers alone do not make a debate rational or informative; they can be false, misleading, used to entrench interests, and so on. Indeed, a key theme in Deringer’s work is that when quantitiative reasoning gains more ground, the question is why, and to whose benefit. In this sense his work aligns with the long-running and always-relevant approach of the Institute’s STS faculty, in thinking carefully about how technology and knowledge is applied to the world.

“The broader culture more has become attuned to STS, whether it’s conversations about AI or algorithmic fairness or climate change or energy, these are simultaneously technical and social issues,” Deringer says. “Teaching undergraduates, I’ve found the awareness of that at MIT has only increased.” For both his research and teaching, Deringer received tenure from MIT earlier this year.

Dig in, work outward

Deringer has been focused on these topics since he was an undergraduate at Harvard University.

“I found myself becoming really interested in the history of economics, the history of practical mathematics, data, statistics, and how it came to be that so much of our world is organized quantitatively,” he says.

Deringer wrote a college thesis about how England measured the land it was seizing from Ireland in the 1600s, and then, after graduating, went to work in the finance sector, which gave him a further chance to think about the application of quantification to modern life.

“That was not what I wanted to do forever, but for some of the conceptual questions I was interested in, the societal life of calculations, I found it to be a really interesting space,” Deringer says.

He returned to academia by pursuing his PhD in the history of science at Princeton University. There, in his first year of graduate school, in the archives, Deringer found 18th-century pamphlets about financial calculations concering the value of stock involved in the infamous episode of speculation known as the South Sea Bubble. That became part of his dissertation; skeptics of the South Sea Bubble were among the prominent early voices bringing data into public debates. It has also helped inform his second book.

First, though, Deringer earned his doctorate from Princeton in 2012, then spent three years as a Mellon Postdoctoral Research Fellow at Columbia University. He joined the MIT faculty in 2015. At the Institute, he finished turning his dissertation into the “Calculated Values” book — which won the 2019 Oscar Kenshur Prize for the best book from the Center for Eighteenth-Century Studies at Indiana University, and was co-winner of the 2021 Joseph J. Spengler Prize for best book from the History of Economics Society.

“My method as a scholar is to dig into the technical details, then work outward historically from them,” Deringer says.

A long historical chain

Even as Deringer was writing his first book, the idea for the second one was taking root in his mind. Those South Sea Bubble pamphets he had found while at Princeton incorporated discounting, which was intermittently present in “Calculated Values.” Deringer was intrigued by how adept 18th-century figures were at discounting.

“Something that I thought of as a very modern technique seemed to be really well-known by a lot of people in the 1720s,” he says.

At the same time, a conversation with an academic colleague in philosophy made it clear to Deringer how different conclusions about discounting had become debated in climate change policy. He soon resolved to write the “biography of a calculation” about financial discounting.

“I knew my next book had to be about this,” Deringer says. “I was very interested in the deep historical roots of discounting, and it has a lot of present urgency.”

Deringer says the book will incorporate material about the financing of English cathedrals, the heavy use of discounting in the mining industry during the Industrial Revolution, a revival of discounting in 1960s policy circles, and climate change, among other things. In each case, he is carefully looking at the interests and historical dynamics behind the use of discounting.

“For people who use discounting regularly, it’s like gravity: It’s very obvious that to be rational is to discount the future according to this formula,” Deringer says. “But if you look at history, what is thought of as rational is part of a very long historical chain of people applying this calculation in various ways, and over time that’s just how things are done. I’m really interested in pulling apart that idea that this is a sort of timeless rational calculation, as opposed to a product of this interesting history.”

Working in STS, Deringer notes, has helped encourage him to link together numerous historical time periods into one book about the numerous ways discounting has been used.

“I’m not sure that pursuing a book that stretches from the 17th century to the 21st century is something I would have done in other contexts,” Deringer says. He is also quick to credit his colleagues in STS and in other programs for helping create the scholarly environment in which he is thriving.

“I came in with a really amazing cohort of other scholars in SHASS,” Deringer notes, referring to the MIT School of Humanities, Arts, and Social Sciences. He cites others receiving tenure in the last year such as his STS colleague Robin Scheffler, historian Megan Black, and historian Caley Horan, with whom Deringer has taught graduate classes on the concept of risk in history. In all, Deringer says, the Institute has been an excellent place for him to pursue interdisciplinary work on technical thought in history.

“I work on very old things and very technical things,” Deringer says. “But I’ve found a wonderful welcoming at MIT from people in different fields who light up when they hear what I’m interested in.”

0 notes

jcmarchi · 2 days

Text

Mini-Gemini: Mining the Potential of Multi-modality Vision Language Models

New Post has been published on https://thedigitalinsider.com/mini-gemini-mining-the-potential-of-multi-modality-vision-language-models/

Mini-Gemini: Mining the Potential of Multi-modality Vision Language Models

The advancements in large language models have significantly accelerated the development of natural language processing, or NLP. The introduction of the transformer framework proved to be a milestone, facilitating the development of a new wave of language models, including OPT and BERT, which exhibit profound linguistic understanding. Furthermore, the inception of GPT, or Generative Pre-trained Transformer models, introduced a new paradigm with autoregressive modeling and established a robust method for language prediction and generation. The advent of language models like GPT-4, ChatGPT, Mixtral, LLaMA, and others has further fueled rapid evolution, with each model demonstrating enhanced performance in tasks involving complex language processing. Among existing methods, instruction tuning has emerged as a key technique for refining the output of pre-trained large language models, and the integration of these models with specific tools for visual tasks has highlighted their adaptability and opened doors for future applications. These extend far beyond the traditional text-based processing of LLMs to include multimodal interactions.

Furthermore, the convergence of natural language processing and computer vision models has given rise to VLMs, or Vision Language Models, which combine linguistic and vision models to achieve cross-modal comprehension and reasoning capabilities. The integration and advent of visual and linguistic models have played a crucial role in advancing tasks that require both language processing and visual understanding. The emergence of revolutionary models like CLIP has further bridged the gap between vision tasks and language models, demonstrating the feasibility and practicality of cross-modal applications. More recent frameworks like LLaMA and BLIP leverage tailored instruction data to devise efficient strategies that demonstrate the potent capabilities of the model. Additionally, combining large language models with image outputs is the focus of recent multimodal research, with recent methods being able to bypass direct generation by utilizing the image retrieval approach to produce image outputs and interleaved texts.

With that being said, and despite the rapid advancements in vision language models facilitating basic reasoning and visual dialogue, there still exists a significant performance gap between advanced models like GPT-4, and vision language models. Mini-Gemini is an attempt to narrow the gap that exists between vision language models and more advanced models by mining the potential of VLMs for better performance from three aspects: VLM-guided generation, high-quality data, and high-resolution visual tokens. To enhance visual tokens, the Mini-Gemini framework proposes to utilize an additional visual encoder for high-resolution refinement without increasing the count of visual tokens. The Mini-Gemini framework further constructs a high-quality dataset in an attempt to promote precise comprehension of images and reasoning-based generation. Overall, the Mini-Gemini framework attempts to mine the potential of vision language models, and aims to empower existing frameworks with image reasoning, understanding, and generative capabilities simultaneously. This article aims to cover the Mini-Gemini framework in depth, and we explore the mechanism, the methodology, the architecture of the framework along with its comparison with state of the art frameworks. So let’s get started.

Over the years, large language models have evolved, and they now boast of remarkable multi-modal capabilities, and are becoming an essential part of current vision language models. However, there exists a gap between the multi-modal performance of large language models and vision language models with recent research looking for ways to combine vision with large language models using images and videos. For vision tasks itself, image resolution is a crucial element to explicitly despite the surrounding environment with minimal visual hallucinations. To bridge the gap, researchers are developing models to improve the visual understanding in current vision language models, and two of the most common approaches are: increasing the resolution, and increasing the number of visual tokens. Although increasing the number of visual tokens with higher resolution images does enhance the visual understanding, the boost is often accompanied with increased computational requirements and associated costs especially when processing multiple images. Furthermore, the capabilities of existing models, quality of existing data, and applicability remains inadequate for an accelerated development process, leaving researchers with the question, “how to accelerate the development of vision language models with acceptable costs”?

The Mini-Gemini framework is an attempt to answer the question as it attempts to explore the potential of vision language models from three aspects: VLM-guided generation or expanded applications, high-quality data, and high-resolution visual tokens. First, the Mini-Gemini framework implements a ConvNet architecture to generate higher-resolution candidates efficiently, enhancing visual details while maintaining the visual token counts for the large language model. The Mini-Gemini framework amalgamates publicly available high-quality datasets in an attempt to enhance the quality of the data, and integrates these enhancements with state of the art generative and large language models with an attempt to enhance the performance of the VLMs, and improve the user experience. The multifaceted strategy implemented by the Mini-Gemini framework enables it to explore hidden capabilities of vision language models, and achieves significant advancements with evident resource constraints.

In general, the Mini-Gemini framework employs an any to any paradigm since it is capable of handling both text and images as input and output. In particular, the Mini-Gemini framework introduces an efficient pipeline for enhancing visual tokens for input images, and features a dual-encoder system comprising of twin encoders: the first encoder is for high-resolution images, while the second encoder is for low-quality visual embedding. During inference, the encoders work in an attention mechanism, where the low-resolution encoder generates visual queries, while the high-resolution encoder provides key and values for reference. To augment the data quality, the Mini-Gemini framework collects and produces more data based on public resources, including task-oriented instructions, generation-related data, and high-resolution responses, with the increased amount and enhanced quality improving the overall performance and capabilities of the model. Furthermore, the Mini-Gemini framework supports concurrent text and image generation as a result of the integration of the vision language model with advanced generative models.

Mini-Gemini : Methodology and Architecture

At its core, the Mini-Gemini framework is conceptually simple, and comprises three components.

The framework employs dual vision encoders to provide low-resolution visual embeddings and high resolution candidates.

The framework proposes to implement patch info mining to conduct mining at patch level between low-resolution visual queries, and high-resolution regions.

The Mini-Gemini framework utilizes a large language model to marry text with images for both generation and comprehension simultaneously.

Dual-Vision Encoders

The Mini-Gemini framework can process both text and image inputs, with the option to handle them either individually or in a combination. As demonstrated in the following image, the Mini-Gemini framework starts the process by employing bilinear interpolation to generate a low-resolution image from its corresponding high-resolution image.

The framework then processes these images and encodes them into a multi-grid visual embedding in two parallel image flows. More specifically, the Mini-Gemini framework maintains the traditional pipeline for low-resolution flows and employs a CLIP-pretrained Visual Transformer to encode the visual embeddings, facilitating the model to preserve the long-range relation between visual patches for subsequent interactions in large language models. For the high-resolution flows, the Mini-Gemini framework adopts the CNN or Convolution Neural Networks based encoder for adaptive and efficient high resolution image processing.

Patch Info Mining

With the dual vision encoders generating the LR embeddings and HR features, the Mini-Gemini framework proposes to implement patch info mining with the aim of extending the potential of vision language models with enhanced visual tokens. In order to maintain the number of visual tokens for efficiency in large language models, the Mini-Gemini framework takes the low-resolution visual embeddings as the query, and aims to retrieve relevant visual cues from the HR feature candidates, with the framework taking the HR feature map as the key and value.

As demonstrated in the above image, the formula encapsulates the process of refining and synthesizing visual cues, which leads to the generation of advanced visual tokens for the subsequent large language model processing. The process ensures that the framework is able to confine the mining for each query to its corresponding sub region in the HR feature map with the pixel-wise feature count, resulting in enhanced efficiency. Owing to this design, the Mini-Gemini framework is able to extract the HR feature details without enhancing the count of visual tokens, and maintains a balance between computational feasibility and richness of detail.

Text and Image Generation

The Mini-Gemini framework concatenates the visual tokens and input text tokens as the input to the large language models for auto-regressive generation. Unlike traditional vision language models, the Mini-Gemini framework supports text-only as well as text-image generation as input and output, i.e. any to any inference, and it is the result of this outstanding image-text understanding and reasoning capabilities, the Mini-Gemini is able to generate high quality images. Unlike recent works that focus on the domain gap between text embeddings of the generation models and large language models, the Mini-Gemini framework attempts to optimize the gap in the domain of language prompts by translating user instructions into high quality prompts that produce context relevant images in latent diffusion models. Furthermore, for a better understanding of instruction finetuning, and cross modality alignment, the Mini-Gemini framework collects samples from publicly available high quality datasets, and uses the GPT-4 turbo framework to further construct a 13K instruction following dataset to support image generation.

Mini-Gemini : Experiments and Results

To evaluate its performance, the Mini-Gemini framework is instantiated with the pre-trained ConvNext-L framework for the HR vision encoder, and with a CLIP-pre-trained Vision Transformer for the LR vision encoder. To ensure training efficiency, the Mini-Gemini framework keeps the two vision encoders fixed, and optimizes the projectors of patch info mining in all stages, and optimizes the large language model during the instruction tuning stage itself.

The following table compares the performance of the Mini-Gemini framework against state of the art models across different settings, and also takes in consideration private models. As it can be observed, the Mini-Gemini outperforms existing frameworks across a wide range of LLMs consistently at normal resolution, and demonstrates superior performance when configured with the Gemma-2B in the category of efficient models. Furthermore, when larger large language models are employed, the scalability of the Mini-Gemini framework is evident.

To evaluate its performance on high resolution and extended visual tokens, the experiments are performed with an input size of 672 for the LR vision encoder, and 1536 for the visual encoder. As mentioned earlier, the main purpose of the HR visual encoder is to offer high-resolution candidate information. As it can be observed, the Mini-Gemini framework delivers superior performance when compared against state of the art frameworks.

Furthermore, to assess the visual comprehension prowess of the Mini-Gemini framework in real-world settings, developers apply the model to a variety of reasoning and understanding tasks as demonstrated in the following image. As it can be observed, the Mini-Gemini framework is able to solve a wide array of complex tasks thanks to the implementation of patch info mining, and high-quality data. But what’s more impressive is the fact that the Mini-Gemini framework demonstrates a keen addition to detail that extends beyond mere recognition prowess, and describes intricate elements intricately.

The following figure provides a comprehensive evaluation of the generative abilities of the Mini-Gemini framework.

When compared against recent models like ChatIllusion and AnyGPT, the Mini-Gemini framework demonstrates stronger multi-modal understanding abilities, allowing it to generate text to image captions that align with the input instructions better, and results in image to text answers with stronger conceptual similarity. What’s more impressive is the fact that the Mini-Gemini framework demonstrates remarkable proficiency in generating high-quality content using multi-model human instructions only with text training data, a capability that illustrates Mini-Gemini’s robust semantic interpretation and image-text alignment skills.

Final Thoughts

In this article we have talked about Mini-Gemini, a potent and streamlined framework for multi-modality vision language models. The primary aim of the Mini-Gemini framework is to harness the latent capabilities of vision language models using high quality data, strategic design of the framework, and an expanded functional scope. Mini-Gemini is an attempt to narrow the gap that exists between vision language models and more advanced models by mining the potential of VLMs for better performance from three aspects: VLM-guided generation, high-quality data, and high-resolution visual tokens. To enhance visual tokens, the Mini-Gemini framework proposes to utilize an additional visual encoder for high-resolution refinement without increasing the count of visual tokens. The Mini-Gemini framework further constructs a high-quality dataset in an attempt to promote precise comprehension of images and reasoning-based generation. Overall, the Mini-Gemini framework attempts to mine the potential of vision language models, and aims to empower existing frameworks with image reasoning, understanding, and generative capabilities simultaneously.

0 notes

jcmarchi · 2 days

Text

Three from MIT awarded 2024 Guggenheim Fellowships

New Post has been published on https://thedigitalinsider.com/three-from-mit-awarded-2024-guggenheim-fellowships/

Three from MIT awarded 2024 Guggenheim Fellowships

MIT faculty members Roger Levy, Tracy Slatyer, and Martin Wainwright are among 188 scientists, artists, and scholars awarded 2024 fellowships from the John Simon Guggenheim Memorial Foundation. Working across 52 disciplines, the fellows were selected from almost 3,000 applicants for “prior career achievement and exceptional promise.”

Each fellow receives a monetary stipend to pursue independent work at the highest level. Since its founding in 1925, the Guggenheim Foundation has awarded over $400 million in fellowships to more than 19,000 fellows. This year, MIT professors were recognized in the categories of neuroscience, physics, and data science.

Roger Levy is a professor in the Department of Brain and Cognitive Sciences. Combining computational modeling of large datasets with psycholinguistic experimentation, his work furthers our understanding of the cognitive underpinning of language processing, and helps to design models and algorithms that will allow machines to process human language. He is a recipient of the Alfred P. Sloan Research Fellowship, the NSF Faculty Early Career Development (CAREER) Award, and a fellowship at the Center for Advanced Study in the Behavioral Sciences.

Tracy Slatyer is a professor in the Department of Physics as well as the Center for Theoretical Physics in the MIT Laboratory for Nuclear Science and the MIT Kavli Institute for Astrophysics and Space Research. Her research focuses on dark matter — novel theoretical models, predicting observable signals, and analysis of astrophysical and cosmological datasets. She was a co-discoverer of the giant gamma-ray structures known as the “Fermi Bubbles” erupting from the center of the Milky Way, for which she received the New Horizons in Physics Prize in 2021. She is also a recipient of a Simons Investigator Award and Presidential Early Career Awards for Scientists and Engineers.

Martin Wainwright is the Cecil H. Green Professor in Electrical Engineering and Computer Science and Mathematics, and affiliated with the Laboratory for Information and Decision Systems and Statistics and Data Science Center. He is interested in statistics, machine learning, information theory, and optimization. Wainwright has been recognized with an Alfred P. Sloan Foundation Fellowship, the Medallion Lectureship and Award from the Institute of Mathematical Statistics, and the COPSS Presidents’ Award from the Joint Statistical Societies. Wainwright has also co-authored books on graphical and statistical modeling, and solo-authored a book on high dimensional statistics.

“Humanity faces some profound existential challenges,” says Edward Hirsch, president of the foundation. “The Guggenheim Fellowship is a life-changing recognition. It’s a celebrated investment into the lives and careers of distinguished artists, scholars, scientists, writers and other cultural visionaries who are meeting these challenges head-on and generating new possibilities and pathways across the broader culture as they do so.”

0 notes

jcmarchi · 2 days

Text

Zero Trust strategies for navigating IoT/OT security challenges - CyberTalk

New Post has been published on https://thedigitalinsider.com/zero-trust-strategies-for-navigating-iot-ot-security-challenges-cybertalk/

Zero Trust strategies for navigating IoT/OT security challenges - CyberTalk

Travais ‘Tee’ Sookoo leverages his 25 years of experience in network security, risk management, and architecture to help businesses of all sizes, from startups to multi-nationals, improve their security posture. He has a proven track record of leading and collaborating with security teams and designing secure solutions for diverse industries.

Currently, Tee serves as a Security Engineer for Check Point, covering the Caribbean region. He advises clients on proactive risk mitigation strategies. He thrives on learning from every challenge and is always looking for ways to contribute to a strong cyber security culture within organizations.

In this informative interview, expert Travais Sookoo shares insights into why organizations need to adopt a zero trust strategy for IoT and how to do so effectively. Don’t miss this!

For our less technical readers, why would organizations want to implement zero trust for IoT systems? What is the value? What trends are you seeing?

For a moment, envision your organization as a bustling apartment building. There are tenants (users), deliveries (data), and of course, all sorts of fancy gadgets (IoT devices). In the old days, our threat prevention capabilities might have involved just a single key for the building’s front door (the network perimeter). Anyone with that key could access everything; the mailbox, deliveries, gadgets.

That’s how traditional security for some IoT systems worked. Once the key was obtained, anyone could gain access. With zero trust, instead of giving everyone the master key, the application of zero trust verifies each device and user ahead of provisioning access.

The world is getting more connected, and the number of IoT devices is exploding, meaning more potential security gaps. Organizations are realizing that zero trust is a proactive way to stay ahead of the curve and keep their data and systems safe.

Zero trust also enables organizations to satisfy many of their compliance requirements and to quickly adapt to ever-increasing industry regulations.

What challenges are organizations experiencing in implementing zero trust for IoT/OT systems?

While zero trust is a powerful security framework, the biggest hurdle I hear about is technology and personnel.

In terms of technology, the sheer number and variety of IoT devices can be overwhelming. Enforcing strong security measures with active monitoring across this diverse landscape is not an easy task. Additionally, many of these devices lack the processing power to run security or monitoring software, thus making traditional solutions impractical.

Furthermore, scaling zero trust to manage the identities and access controls for potentially hundreds, thousands, even millions of devices can be daunting.

Perhaps the biggest challenge is that business OT systems must prioritize uptime and reliability above all else. Implementing zero trust may require downtime or potentially introduce new points of failure. Finding ways to achieve zero trust without compromising the availability of critical systems takes some manoeuvring.

And now the people aspect: Implementing and maintaining a zero trust architecture requires specialized cyber security expertise, which many organizations may not have. The talent pool for these specialized roles can be limited, making it challenging to recruit and retain qualified personnel.

Additionally, zero trust can significantly change how people interact with OT systems. Organizations need to invest in training staff on new procedures and workflows to ensure a smooth transition.

Could you speak to the role of micro-segmentation in implementing zero trust for IoT/OT systems? How does it help limit lateral movement and reduce the attack surface?

With micro-segmentation, we create firewalls/access controls between zones, making it much harder for attackers to move around. We’re locking the doors between each room in the apartment; even if an attacker gets into the thermostat room (zone), they can’t easily access the room with our valuables (critical systems).

The fewer devices and systems that an attacker can potentially exploit, the better. Micro-segmentation reduces the overall attack surface and the potential blast radius by limiting what devices can access on the network.

Based on your research and experience, what are some best practices or lessons learned in implementing zero trust for IoT and OT systems that you can share with CISOs?

From discussions I’ve had and my research:

My top recommendation is to understand the device landscape. What are the assets you have, their purpose, how critical are they to the business? By knowing the environment, organizations can tailor zero trust policies to optimize both security and business continuity.

Don’t try to boil the ocean! Zero trust is a journey, not a destination. Start small, segmenting critical systems and data first. Learn from that experience and then expand the implementation to ensure greater success with declining margins of errors.

Legacy OT systems definitely throw a wrench into plans and can significantly slow adoption of zero trust. Explore how to integrate zero trust principles without compromising core functionalities. It might involve a mix of upgrades and workarounds.

The core principle of zero trust is granting only the minimum access required for a device or user to function (least privilege). Document who needs what and then implement granular access controls to minimize damage from a compromised device.

Continuous monitoring of network activity and device behaviour is essential to identify suspicious activity and potential breaches early on. Ensure that monitoring tools encompasses everything and your teams can expertly use it.

Automating tasks, such as device onboarding, access control enforcement, and security patching can significantly reduce the burden on security teams and improve overall efficiency.

Mandate regular review and policy updates based on new threats, business needs, and regulatory changes.

Securing IoT/OT systems also requires close collaboration between OT and IT teams. Foster teamwork, effective communications and understanding between these departments to break down silos. This cannot be stressed enough. Too often, the security team is the last to weigh in, often after it’s too late.

What role can automation play in implementing and maintaining Zero Trust for IoT/OT systems?

Zero trust relies on granting least privilege access. Automation allows us to enforce these granular controls by dynamically adjusting permissions based on device type, user role, and real-time context.

Adding new IoT devices can be a tedious process and more so if there are hundreds or thousands of these devices. However, automation can greatly streamline device discovery, initial configuration, and policy assignment tasks, thereby freeing up security teams to focus on more strategic initiatives.

Manually monitoring a complex network with numerous devices is overwhelming, but we can automate processes to continuously monitor network activity, device behaviour, and identify anomalies that might indicate a potential breach. And if a security incident occurs, we can automate tasks to isolate compromised devices, notifying security teams, and initiating remediation procedures.

Through monitoring, it’s possible to identify IoT/OT devices that require patching, which can be crucial, but also time-consuming. It’s possible to automate patch deployment with subsequent verification, and even launch rollbacks in case of unforeseen issues.

If this sounds as a sales pitch, then hopefully you’re sold. There’s no doubt that automation will significantly reduce the burden on security teams, improve the efficiency of zero trust implementation and greatly increase our overall security posture.

What metrics would you recommend for measuring the effectiveness of zero trust implementation in IoT and OT environments?

A core tenet of zero trust is limiting how attackers move between devices or otherwise engage in lateral movement. The number of attempted lateral movements detected and blocked can indicate the effectiveness of segmentation and access controls.

While some breaches are inevitable, a significant decrease in compromised devices after implementing zero trust signifies a positive impact. This metric should be tracked alongside the severity of breaches and the time it takes to identify and contain them. With zero trust, it is assumed any device or user, regardless of location, could be compromised.

The Mean Time to Detection (MTD) and Mean Time to Response (MTTR) are metrics that you can use to measure how quickly a security incident is identified and contained. Ideally, zero trust should lead to faster detection and response times, minimizing potential damage.

Zero trust policies enforces granular access controls. Tracking the number of least privilege violations (users or devices accessing unauthorized resources) can expose weaknesses in policy configuration or user behaviour and indicate areas for improvement.

Security hygiene posture goes beyond just devices. It includes factors like patch compliance rates, and the effectiveness of user access.

Remember the user experience? Tracking user satisfaction with the zero trust implementation process and ongoing security measures can help identify areas for improvement and ensure a balance between security and usability.

It’s important to remember that zero trust is a journey, not a destination. The goal is to continuously improve our security posture and make it more difficult for attackers to exploit vulnerabilities in our IoT/OT systems. Regularly review your metrics and adjust zero trust strategies as needed.

Is there anything else that you would like to share with the CyberTalk.org audience?

Absolutely! As we wrap up this conversation, I want to leave the CyberTalk.org audience with a few key takeaways concerning securing IoT and OT systems:

Zero trust is a proactive approach to security. By implementing zero trust principles, organizations can significantly reduce the risk of breaches and protect their critical infrastructure.

Don’t go it alone: Security is a team effort. Foster collaboration between IT, OT, and security teams to ensure that everyone is on the same page when it comes to adopting zero trust.

Keep learning: The cyber security landscape is constantly evolving. Stay up-to-date on the latest threats and best practices. Resources like Cybertalk.org are a fantastic place to start.

Focus on what matters: A successful zero trust implementation requires a focus on all three pillars: people, process, and technology. Security awareness training for employees, clearly defined policies and procedures, and the right security tools are all essential elements.

Help is on the way: Artificial intelligence and machine learning will play an increasingly important role in automating zero trust processes and making them even more effective.

Thank you, CyberTalk.org, for the opportunity to share my thoughts. For more zero trust insights, click here.

0 notes

jcmarchi · 2 days

Text

A musical life: Carlos Prieto ’59 in conversation and concert

New Post has been published on https://thedigitalinsider.com/a-musical-life-carlos-prieto-59-in-conversation-and-concert/

A musical life: Carlos Prieto ’59 in conversation and concert

World-renowned cellist Carlos Prieto ’59 returned to campus for an event to perform and to discuss his new memoir, “Mi Vida Musical.”

At the April 9 event in the Samberg Conference Center, Prieto spoke about his formative years at MIT and his subsequent career as a professional cellist. The talk was followed by performances of J.S. Bach’s “Cello Suite No. 3” and Eugenio “Toussaint’s Bachriation.” Valerie Chen, a 2022 Sudler Prize winner and Emerson/Harris Fellow, also performed Phillip Glass’s “Orbit.”

Prieto was born in Mexico City and began studying the cello when he was 4. He graduated from MIT with BS degrees in 1959 in Course 3, then called the Metallurgical Engineering and today Materials Science and Engineering, and in Course 14 (Economics). He was the first cello and soloist of the MIT Symphony Orchestra. While at MIT, he took all available courses in Russian, which allowed him, years later, to study at Lomonosov University in Moscow.

After graduation from MIT, Prieto returned to Mexico, where he rose to become the head of an integrated iron and steel company.

“When I returned to Mexico, I was very active in my business life, but I was also very active in my music life,” he told the audience. “And at one moment, the music overcame all the other activities and I left my business activities to devote all my time to the cello and I’ve been doing this for the past 50 years.”

During his musical career, Prieto played all over the world and has played and recorded the world premieres of 115 compositions, most of which were written for him. He is the author of 14 books, some of which have been translated into English, Russian, and Portuguese.

Prieto’s honors include the Order of the Arts and Letters from France, the Order of Civil Merit from the King of Spain, and the National Prize for Arts and Sciences from the president of Mexico. In 1993 he was appointed member of the MIT Music and Theater Advisory Committee. In 2014, the School of Humanities, Arts, and Social Sciences awarded Prieto the Robert A. Muh Alumni Award.

0 notes

jcmarchi · 2 days

Text

TopSpin 2K25 Review - A Strong Return - Game Informer

New Post has been published on https://thedigitalinsider.com/topspin-2k25-review-a-strong-return-game-informer/

TopSpin 2K25 Review - A Strong Return - Game Informer

In the heyday of the tennis-sim video game genre, Top Spin and Virtua Tennis were the best players in the crowded space. However, in the time since the genre’s boom settled, the offerings have fallen off considerably, with both franchises going more than a decade without a new release. TopSpin 2K25 signals the reemergence of the critically acclaimed series, and though it’s been a while since it stepped on the court, it’s evident the franchise hasn’t lost its stroke.

TopSpin 2K25 faithfully recreates the high-speed chess game of real-world tennis. Positioning, spin, timing, and angles are critical to your success. For those unfamiliar with those fundamental tennis tenets, 2K25 does a superb job of onboarding players with TopSpin Academy, which covers everything from where you should stand to how to play different styles. Even as someone who played years of tennis in both real life and video games, I enjoyed going through the more advanced lessons to refamiliarize myself with the various strategies at play.

Once on the court, you learn how crucial those tactics are. The margin of error is extremely thin, as the difference between a winner down the baseline and a shot into the net is often a split-second on the new timing meter. This meter ensures you release the stroke button timed with when the ball is in the ideal striking position relative to your player. Mastering this is pivotal, as it not only improves your shot accuracy but also your power.

[embedded content]

TopSpin 2K25 is at its best when you’re in sustained rallies against an evenly-matched opponent. Getting off a strong serve to immediately puts your opponent on the defensive, then trying to capitalize on their poor positioning as they struggle to claw back into the point effectively captures the thrill of the real-world game. I also love how distinct each play style feels in action; an offensive baseline player like Serena Williams presents different challenges than a serve-and-volleyer like John McEnroe.

You can hone your skills in one-off exhibition matches, but I spent most of my time in TopSpin 2K25 in MyCareer. Here, you create your player, with whom you’ll train and climb the ranks. As you complete challenges and win matches, you raise your status, which opens new features like upgradeable coaches, equippable skills, and purchasable homes to alleviate the stamina drain from travel. Managing your stamina by sometimes resting is essential to sustain high-level play; pushing yourself too hard can even cause your player to suffer injuries that sideline you for months.

I loved most of my time in MyCareer, but some design decisions ruined the immersion. For example, I ignored portions of the career goals necessary to rank up my player for hours, so while I was in the top 10 global rankings, I was unable to participate in a Grand Slam because I was still at a lower status than my ranking would typically confer. And since repetition is the path to mastery, it’s counterintuitive that repeated training minigames award fewer benefits, particularly since the mode as a whole is a repetitive loop of training, special events, and tournaments. Additionally, MyCareer shines a light on the shallow pool of licensed players on offer. Most of my matches were against created characters, even hours deep. 2K has promised free licensed pros in the post-launch phase, but for now, the game is missing multiple top players.

I’m pleasantly surprised by how unintrusive the use of VC is. In the NBA 2K series, VC, which can be earned slowly or bought using real money, is used to directly improve your player. In TopSpin 2K25, it’s used primarily for side upgrades, like leveling up your coach, relocating your home, earning XP boosts, resetting your attribute distribution, or purchasing cosmetics. Though I’m still not a fan of microtransactions affecting a single-player mode – particularly since it’s almost certainly why you need to be online to play MyCareer – it’s much more palatable than its NBA contemporary.

If you’d rather play against real opponents, you can show off your skills (and your created character) in multiple online modes. World Tour pits your created player against others across the globe in various tournaments and leaderboard challenges, while 2K Tour leverages the roster of licensed players with daily challenges to take on. Outside of minor connection hiccups, I had an enjoyable time tackling the challenges presented by other players online. However, World Tour’s structure means that despite the game’s best efforts, mismatches occur; it’s no fun to play against a created character multiple levels higher than you. Thankfully, these mismatches were the outlier rather than the exception in my experience.

TopSpin 2K25 aptly brings the beloved franchise back to center court, showing that not only does the series still have legs, but so does the sim-tennis genre as a whole. Though its modes are somewhat repetitive and it’s missing several high-profile pros at launch, TopSpin 2K25 serves up a compelling package for tennis fans.

0 notes

jcmarchi · 2 days

Text

1000 AI-powered machines: Vision AI on an industrial scale

New Post has been published on https://thedigitalinsider.com/1000-ai-powered-machines-vision-ai-on-an-industrial-scale/

1000 AI-powered machines: Vision AI on an industrial scale

This article is based on Bart Baekelandt’s brilliant talk at the Computer Vision Summit in London. Pro and Pro+ members can enjoy the complete recording here. For more exclusive content, head to your membership dashboard.

Hi, I’m Bart Baekelandt, Head of Product Management at Robovision.

Today, we’re going to talk about the lessons we’ve learned over the last 15 years of applying AI to robots and machines at scale in the real world.

Robovision’s journey: From flawed to fantastic

First, let’s look at what these machines were like in the past.

15 years ago, our machine was very basic with extremely rudimentary computer vision capabilities. It used classical machine vision techniques and could only handle very basic tasks like recognizing a hand. Everything was hard-coded, so if you needed the machine to recognize something new, you’d have to recode the entire application. It was expensive and required highly skilled personnel.

Nowadays, we don’t have just one machine – we have entire populations of machines, with advanced recognition capabilities. There’s a continuous process of training AI models and applying them to production so the machines can tackle the problem at hand.

For example, there are machines that can take a seedling from a conveyor belt and plant it in a tray. We have entire fleets of these specialized machines. One day they’re trained to handle one type of seedling, and the next day they’re retrained to perform optimally for a different variety of plant.

So yeah, a lot has happened in 15 years. We’ve gone from initially failing to scale AI, to figuring out how to apply AI at scale with minimal support from our side. As of today, we’ve produced over 1000 machines with game-changing industrial applications

Let’s dive into a few of the key lessons we’ve picked up along the way.

Lesson one: AI success happens after the pilot

The first lesson is that AI success happens after the pilot phase. We learned this lesson the hard way in the initial stages of applying AI, around 2012.

Let me share a quick anecdote. When we were working on the machine that takes seedlings from a conveyor belt and plants them in trays, we spent a lot of time applying AI and building the algorithm to recognize the right breaking point on each seedling and plant it properly.

Eventually, we nailed it – the algorithm worked perfectly. The machine builder who integrated it was happy, and the customer growing the seedlings was delighted because everything was functioning as intended.

However, the congratulations were short-lived. Within two weeks, we got a call – the system wasn’t picking the seedlings well anymore. What had happened? They were now trying to handle a different seedling variety, and the images looked just different enough that our AI model struggled. The robot started missing the plants entirely or planting them upside down.

We got new image data from the customer’s operations and retrained the model. Great, it worked again! But sure enough, two weeks later, we got another call reporting the same problem all over again.

This highlighted a key problem. The machine builder wanted to sell to many customers, but we couldn’t feasibly support each one by perpetually retraining models on their unique data. That approach doesn’t scale.

That painful lesson was the genesis of our products. We realized the end customers needed to be able to continuously retrain the models themselves without our assistance. So, we developed tooling for them to capture new data, convert it to retrained models, deploy those models to the machines, and interface with the machines for inference.

Our product philosophy stems directly from those harsh real-world lessons about what’s required to successfully scale AI in real-world production.

Lesson two: It’s about getting the odd couple to work together

When you’re creating working AI solutions at scale, there typically are two types of people involved. They’re your classic “odd couple,” but they need to be able to collaborate effectively.

On one hand, you have the data scientists – they generally have advanced degrees like Masters in Engineering or even PhDs. Data scientists are driven by innovation. They live to solve complex problems and find solutions to new challenges.

Once they’ve cracked the core issue, however, they tend to lose interest. They want to move on to the next big innovation, rather than focusing on continuous improvement cycles or incremental optimizations

On the other hand, you have the machine operators who run the manufacturing systems and processes where AI gets applied at scale – whether that’s a factory, greenhouse, or another facility.

The machine operators have intricate knowledge of the products being handled by the machines. If you’re deploying AI to handle seedlings, for example, no one understands the nuances, variations, and defects of those plants better than the operator.

This post is for paying subscribers only

Subscribe now

Already have an account? Sign in

0 notes

jcmarchi · 2 days

Text

A snapshot of bias, the human mind, and AI

New Post has been published on https://thedigitalinsider.com/a-snapshot-of-bias-the-human-mind-and-ai/

A snapshot of bias, the human mind, and AI

Introducing bias & the human mind

The human mind is a landscape filled with curiosity and, at times irrationality, motivation, confusion, and bias. The latter results in levels of complexity in how both the human and more recently, the artificial slant affects artificial intelligence systems from concept to scale.

Bias is something that in many cases unintentionally appears – whether it be in human decision-making or the dataset – but its impact on output can be sizeable. With several cases over the years highlighting the social, political, technological, and environmental impact of bias, this piece will explore this important topic and some thoughts on how such a phenomenon can be managed.

Whilst there’s many variations and interpretations (which in some cases themselves could be biased), let’s instead of referring to a definition explore how the human mind might work in certain scenarios.

Imagine two friends (friend A and friend B) at school who’ve had a falling out and makeup again after apologies are exchanged. With friend A’s birthday coming up, they’re going through their invite list and land on Person B (who they fell out with).

Do they want to invite them back and risk the awkwardness if another falling out occurs, or should they take the view they should only invite those they’ve always got along with? The twist is though, Person A choosing the attendees for the party may have had minor falling outs with them in the past, but they’re interpreting it through the lens any previous falling outs are insignificant enough to be looked over.

The follow-up from the above example turns to whether person’s A decision is fair. Now, fairness adds to the difficulty as there’s no scientific definition of what fairness really is.

However, some might align fairness with making a balanced judgment after considering the facts or doing what is right (even if that’s biased!). These are just a couple of ways in which the mind can distort, and mould the completion of tasks, whether they’re strategic or technical.

Before going into the underlying ways in which bias can be managed in AI systems, let’s start from the top: leadership.

Leadership, bias, and Human In the Loop Systems

The combination of leadership and bias introduces important discussions about how such a trait can be managed. “The fish rots from the head down” is a common phrase used to describe leadership styles and their impact across both the wider company and their teams, but this phrase can also be extended to how bias weaves down the chain of command.

For example, if a leader within the C-suite doesn’t get along with the CEO or has had several previous tense exchanges, they may ultimately, subconsciously have a blurred view of the company vision that then spills down, with distorted conviction, to the teams.

Leadership and bias will always remain an important conversation in the boardroom, and there’s been some fascinating studies exploring this in more depth, for example, Shaan Madhavji’s piece on the identification and management of leadership bias [1]. It’s an incredibly eye-opening subject, and one that in my view will become increasingly topical as time moves on.

Generative Artificial Intelligence Report 2024

We’re diving deep into the world of generative artificial intelligence with our new report: Generative AI 2024, which will explore how and why companies are (or aren’t) using this technology.

As we shift from leadership styles and bias to addressing bias in artificial intelligence-based systems, an area that’ll come under further spotlight will be the effectiveness of Human In the Loop Systems (HITL).

Whilst their usefulness varies across industries, in summary, HITL systems fuse both the art of human intuition and the efficiency of machines: an incredibly valuable partnership where complex decision-making at speed is concerned.

Additionally, when linked to bias, the human link in the chain can be key in identifying bias early on to ensure adverse effects aren’t felt later on. On the other hand, HITL won’t always be a Spring cleaning companion: complexities around getting a sizeable batch of training data combined with practitioners who can effectively integrate into a HITL environment can blur the productivity vs efficiency drive the company is aiming to achieve.

Conclusions & the future of bias

In my view, irrespective of how much better HITL systems might (or might not) become, I don’t believe bias can be eliminated, and I don’t believe in the future – no matter how advanced and intelligent AI becomes – we’ll be able to get rid of it.

It’s very much something that’s so woven that it’s not always possible to see or even discern it. Furthermore, sometimes bias traits are only revealed when an individual points it out to someone else, and even then there can be bias on top of bias!

As we look to the future of Generative AI, its associated increasingly challenging ethical considerations, and the wide-ranging debate on how far its usefulness will stem at scale, an important thought will always remain at heart: we on occasions won’t be able to mitigate future impacts of bias until we’re right at the moment and the impact is being felt there and then.

Bibliography

[1] shaan-madhavji.medium.com. (n.d.). Leadership Bias: 12 cognitive biases to become a decisive leader. [online] Available at: https://hospitalityinsights.ehl.edu/leadership-bias.

Want to read more from Ana? Check out one of her articles below:

Navigating artificial intelligence in 2024

Discover how businesses can harness AI’s potential, balance innovation with ethics, and tackle the digital skills gap.

0 notes

jcmarchi · 2 days

Text

Blizzard Announces It's Skipping BlizzCon This Year

New Post has been published on https://thedigitalinsider.com/blizzard-announces-its-skipping-blizzcon-this-year/

Blizzard Announces It's Skipping BlizzCon This Year

Blizzard has decided to cancel this year’s BlizzCon. The company states the event will return in the future, but it plans to showcase upcoming games in a different manner over the coming months.

First announced in a blog post, Blizzard plans to share details on upcoming games like World of Warcraft: The War Within and Diablo IV’s Vessel of Hatred expansion at other trade shows, such as Gamescom. The company also plans to launch “multiple, global, in-person events” for Warcraft’s 30th anniversary, which are described as being “distinct” from BlizzCon.

“Our hope is that these experiences – alongside several live-streamed industry events where we’ll keep you up to date with what’s happening in our game universes – will capture the essence of what makes the Blizzard community so special,” Blizzard states in the blog post.

A Blizzard representative tells Windows Central that Blizzard made the call to cancel BlizzCon and not Microsoft, which completed its acquisition of the company last year. In a statement to the outlet, the representative says, “This is a Blizzard decision. We have explored different event formats in the past, and this isn’t the first time we’re skipping BlizzCon or trying something new. While we have great things to share in 2024, the timing just doesn’t line up for one single event at the end of the year.”

BlizzCon began in 2005 as an annual convention celebrating all things Blizzard. Last year’s show saw the reveal of World of Warcraft’s next expansion, The War Within, as well as two other expansions coming after it. It’s good that event is only taking a year off as opposed to being canned for good, and we’re curious to see how the alternative events shape up over the coming months.

1 note · View note

jcmarchi · 2 days

Text

Generative AI’s Role in Job Satisfaction

New Post has been published on https://thedigitalinsider.com/generative-ais-role-in-job-satisfaction/

Generative AI’s Role in Job Satisfaction

Generative AI (GenAI) is a pivotal technology that enhances work in a myriad of ways. From automating complex analysis to simulating scenarios that assist in decision-making, GenAI use cases are making a big impact across a broad swath of industries, including financial services, consultancies, information technology, legal, telecommunication and more.

Certainly, organizations recognize GenAI’s potential with the increasing adoption of AI within organizations. According to a PWC survey, 73% of U.S. companies have adopted AI in some areas of their business. Yet, discussion persists about GenAI’s role within the workplace, given fears over job displacement, bias, decision-making transparency and more. Despite this, GenAI has made AI technology much more accessible to employees within organizations, regardless of their specific roles.

In fact, a LexisNexis Future of Work survey showed that 72% of professionals anticipate a positive impact from GenAI, and only 4% see it as a threat to job security. GenAI can automate mundane tasks, allowing users to focus on more specialized, impactful and strategic tasks. This, in turn, can increase employee productivity and job satisfaction while ensuring human ambition and innovation walk hand in hand.

AI’s Productivity Boost

GenAI’s rapid rise marks a crucial shift in how organizations must operate and strategize to augment every role. GenAI applications are as diverse as they are impactful. It’s not just hype; GenAI is already poised to increase labor productivity by 0.1 to 0.6% annually through 2040.

GenAI has also created value across multiple sectors and industries. Significant business functions, including Sales, Marketing, Customer Operations and Technology have leveraged GenAI to increase productivity. In technology, for example, GenAI-based coding assistants are a massive help to software developers in suggesting code snippets, refactoring code, fixing bugs, understanding complex code, writing unit tests, documentation and creating complete end-to-end applications.

As employees experiment and explore with GenAI tools, their comfort level with the technology increases. Eighty-six percent of professionals ‘agree’ or ‘strongly agree’ with a willingness to embrace GenAI for both creative and professional work. Sixty-eight percent of employees plan to use GenAI tools for work purposes, while 69% are already using these tools to assist with daily tasks. The data makes it clear that organizations that adopt GenAI can boost productivity, and employees are willing to use it to accelerate efficiency.

Productivity Gains Are a Given, But Also AI Helps with Job Satisfaction

One of the most significant opportunities around GenAI lies in its power to help with job satisfaction. While professionals have fairly balanced expectations on how far adoption will go, 82% expect generative AI to take over a range of repetitive administrative tasks by automating routine tasks and data analysis, freeing them to focus on more strategic aspects of their work.

When asked how they perceive GenAI’s role in the work environment, more than two-thirds of professionals see it as a ‘helpful tool’ or ‘supportive co-worker.’ As a result, they recognize AI’s potential to enhance, not hinder, job performance and are embracing it with a positive mindset toward eliminating repetitive tasks and freeing up time for more rewarding, higher-value work.

Most professionals do not see generative AI as a detriment to job satisfaction, either. Over half (51%) say job satisfaction has improved significantly or moderately thanks to GenAI, while only 10% felt that it decreases job satisfaction. A fundamental rethink is necessary where and how organizations implement GenAI tools within the workplace.

Recommendations to Improve Engagement and Job Satisfaction

Organizations need to consider employee engagement throughout the adoption process of GenAI tools. Here are some recommendations to improve engagement and thereby increase job satisfaction:

Engage your employees to identify the use cases that are most impactful for a particular role or group. Pick tasks that are most time-consuming and tedious, such that solving them would free up time to focus on more critical items.

Identify the GenAI tools and large language models (LLMs) that are most effective for solving the identified use case. Take the time to experiment, test and validate the output. Ensure that you account for a diverse set of inputs for the use case and measure the output quality, including the hallucination rate, to help build trust within your employee base using the solution.

Provide training to your team. Take advantage of the vast information available on the web, with videos, code samples, tool vendor resources and tutorials on using the specific tool, LLM, associated prompts and guardrails. Create mentors and experts within the team to help coach the rest. Showcase examples of lessons learned and success stories to inspire team members who may not see the value.

Identify and measure KPIs. These could include adoption, productivity gains, costs saved or repurposed, employee satisfaction, quality improvement and other KPIs that may be specific to the team or business.

Gen AI isn’t just for technologists anymore; it’s making potent tools accessible to everyone. Most business professionals who once viewed these technologies with skepticism now accept and even welcome them. And it’s no secret why, given GenAI’s power to present organizations and employees alike with unprecedented opportunities toward the future of work.

0 notes

jcmarchi · 2 days

Text

Decoder-Based Large Language Models: A Complete Guide

New Post has been published on https://thedigitalinsider.com/decoder-based-large-language-models-a-complete-guide/

Decoder-Based Large Language Models: A Complete Guide

Large Language Models (LLMs) have revolutionized the field of natural language processing (NLP) by demonstrating remarkable capabilities in generating human-like text, answering questions, and assisting with a wide range of language-related tasks. At the core of these powerful models lies the decoder-only transformer architecture, a variant of the original transformer architecture proposed in the seminal paper “Attention is All You Need” by Vaswani et al.

In this comprehensive guide, we will explore the inner workings of decoder-based LLMs, delving into the fundamental building blocks, architectural innovations, and implementation details that have propelled these models to the forefront of NLP research and applications.

The Transformer Architecture: A Refresher

Before diving into the specifics of decoder-based LLMs, it’s essential to revisit the transformer architecture, the foundation upon which these models are built. The transformer introduced a novel approach to sequence modeling, relying solely on attention mechanisms to capture long-range dependencies in the data, without the need for recurrent or convolutional layers.

The original transformer architecture consists of two main components: an encoder and a decoder. The encoder processes the input sequence and generates a contextualized representation, which is then consumed by the decoder to produce the output sequence. This architecture was initially designed for machine translation tasks, where the encoder processes the input sentence in the source language, and the decoder generates the corresponding sentence in the target language.

Self-Attention: The Key to Transformer’s Success

At the heart of the transformer lies the self-attention mechanism, a powerful technique that allows the model to weigh and aggregate information from different positions in the input sequence. Unlike traditional sequence models, which process input tokens sequentially, self-attention enables the model to capture dependencies between any pair of tokens, regardless of their position in the sequence.

The self-attention operation can be broken down into three main steps:

Query, Key, and Value Projections: The input sequence is projected into three separate representations: queries (Q), keys (K), and values (V). These projections are obtained by multiplying the input with learned weight matrices.

Attention Score Computation: For each position in the input sequence, attention scores are computed by taking the dot product between the corresponding query vector and all key vectors. These scores represent the relevance of each position to the current position being processed.

Weighted Sum of Values: The attention scores are normalized using a softmax function, and the resulting attention weights are used to compute a weighted sum of the value vectors, producing the output representation for the current position.

Multi-head attention, a variant of the self-attention mechanism, allows the model to capture different types of relationships by computing attention scores across multiple “heads” in parallel, each with its own set of query, key, and value projections.

Architectural Variants and Configurations

While the core principles of decoder-based LLMs remain consistent, researchers have explored various architectural variants and configurations to improve performance, efficiency, and generalization capabilities. In this section, we’ll delve into the different architectural choices and their implications.

Architecture Types

Decoder-based LLMs can be broadly classified into three main types: encoder-decoder, causal decoder, and prefix decoder. Each architecture type exhibits distinct attention patterns, as illustrated in Figure 1.

Encoder-Decoder Architecture

Based on the vanilla Transformer model, the encoder-decoder architecture consists of two stacks: an encoder and a decoder. The encoder uses stacked multi-head self-attention layers to encode the input sequence and generate latent representations. The decoder then performs cross-attention on these representations to generate the target sequence. While effective in various NLP tasks, few LLMs, such as Flan-T5, adopt this architecture.

Causal Decoder Architecture

The causal decoder architecture incorporates a unidirectional attention mask, allowing each input token to attend only to past tokens and itself. Both input and output tokens are processed within the same decoder. Notable models like GPT-1, GPT-2, and GPT-3 are built on this architecture, with GPT-3 showcasing remarkable in-context learning capabilities. Many LLMs, including OPT, BLOOM, and Gopher, have widely adopted causal decoders.

Prefix Decoder Architecture

Also known as the non-causal decoder, the prefix decoder architecture modifies the masking mechanism of causal decoders to enable bidirectional attention over prefix tokens and unidirectional attention on generated tokens. Like the encoder-decoder architecture, prefix decoders can encode the prefix sequence bidirectionally and predict output tokens autoregressively using shared parameters. LLMs based on prefix decoders include GLM130B and U-PaLM.

All three architecture types can be extended using the mixture-of-experts (MoE) scaling technique, which sparsely activates a subset of neural network weights for each input. This approach has been employed in models like Switch Transformer and GLaM, with increasing the number of experts or total parameter size showing significant performance improvements.

Decoder-Only Transformer: Embracing the Autoregressive Nature

While the original transformer architecture was designed for sequence-to-sequence tasks like machine translation, many NLP tasks, such as language modeling and text generation, can be framed as autoregressive problems, where the model generates one token at a time, conditioned on the previously generated tokens.

Enter the decoder-only transformer, a simplified variant of the transformer architecture that retains only the decoder component. This architecture is particularly well-suited for autoregressive tasks, as it generates output tokens one by one, leveraging the previously generated tokens as input context.

The key difference between the decoder-only transformer and the original transformer decoder lies in the self-attention mechanism. In the decoder-only setting, the self-attention operation is modified to prevent the model from attending to future tokens, a property known as causality. This is achieved through a technique called “masked self-attention,” where attention scores corresponding to future positions are set to negative infinity, effectively masking them out during the softmax normalization step.

Architectural Components of Decoder-Based LLMs

While the core principles of self-attention and masked self-attention remain the same, modern decoder-based LLMs have introduced several architectural innovations to improve performance, efficiency, and generalization capabilities. Let’s explore some of the key components and techniques employed in state-of-the-art LLMs.

Input Representation

Before processing the input sequence, decoder-based LLMs employ tokenization and embedding techniques to convert the raw text into a numerical representation suitable for the model.

Tokenization: The tokenization process converts the input text into a sequence of tokens, which can be words, subwords, or even individual characters, depending on the tokenization strategy employed. Popular tokenization techniques for LLMs include Byte-Pair Encoding (BPE), SentencePiece, and WordPiece. These methods aim to strike a balance between vocabulary size and representation granularity, allowing the model to handle rare or out-of-vocabulary words effectively.

Token Embeddings: After tokenization, each token is mapped to a dense vector representation called a token embedding. These embeddings are learned during the training process and capture semantic and syntactic relationships between tokens.

Positional Embeddings: Transformer models process the entire input sequence simultaneously, lacking the inherent notion of token positions present in recurrent models. To incorporate positional information, positional embeddings are added to the token embeddings, allowing the model to distinguish between tokens based on their positions in the sequence. Early LLMs used fixed positional embeddings based on sinusoidal functions, while more recent models have explored learnable positional embeddings or alternative positional encoding techniques like rotary positional embeddings.

Multi-Head Attention Blocks

The core building blocks of decoder-based LLMs are multi-head attention layers, which perform the masked self-attention operation described earlier. These layers are stacked multiple times, with each layer attending to the output of the previous layer, allowing the model to capture increasingly complex dependencies and representations.

Attention Heads: Each multi-head attention layer consists of multiple “attention heads,” each with its own set of query, key, and value projections. This allows the model to attend to different aspects of the input simultaneously, capturing diverse relationships and patterns.

Residual Connections and Layer Normalization: To facilitate the training of deep networks and mitigate the vanishing gradient problem, decoder-based LLMs employ residual connections and layer normalization techniques. Residual connections add the input of a layer to its output, allowing gradients to flow more easily during backpropagation. Layer normalization helps to stabilize the activations and gradients, further improving training stability and performance.

Feed-Forward Layers

In addition to multi-head attention layers, decoder-based LLMs incorporate feed-forward layers, which apply a simple feed-forward neural network to each position in the sequence. These layers introduce non-linearities and enable the model to learn more complex representations.

Activation Functions: The choice of activation function in the feed-forward layers can significantly impact the model’s performance. While earlier LLMs relied on the widely-used ReLU activation, more recent models have adopted more sophisticated activation functions like the Gaussian Error Linear Unit (GELU) or the SwiGLU activation, which have shown improved performance.

Sparse Attention and Efficient Transformers

While the self-attention mechanism is powerful, it comes with a quadratic computational complexity with respect to the sequence length, making it computationally expensive for long sequences. To address this challenge, several techniques have been proposed to reduce the computational and memory requirements of self-attention, enabling efficient processing of longer sequences.

Sparse Attention: Sparse attention techniques, such as the one employed in the GPT-3 model, selectively attend to a subset of positions in the input sequence, rather than computing attention scores for all positions. This can significantly reduce the computational complexity while maintaining reasonable performance.

Sliding Window Attention: Introduced in the Mistral 7B model , sliding window attention (SWA) is a simple yet effective technique that restricts the attention span of each token to a fixed window size. This approach leverages the ability of transformer layers to transmit information across multiple layers, effectively increasing the attention span without the quadratic complexity of full self-attention.

Rolling Buffer Cache: To further reduce memory requirements, especially for long sequences, the Mistral 7B model employs a rolling buffer cache. This technique stores and reuses the computed key and value vectors for a fixed window size, avoiding redundant computations and minimizing memory usage.

Grouped Query Attention: Introduced in the LLaMA 2 model, grouped query attention (GQA) is a variant of the multi-query attention mechanism that divides attention heads into groups, each group sharing a common key and value matrix. This approach strikes a balance between the efficiency of multi-query attention and the performance of standard self-attention, providing improved inference times while maintaining high-quality results.

Model Size and Scaling

One of the defining characteristics of modern LLMs is their sheer scale, with the number of parameters ranging from billions to hundreds of billions. Increasing the model size has been a crucial factor in achieving state-of-the-art performance, as larger models can capture more complex patterns and relationships in the data.

Parameter Count: The number of parameters in a decoder-based LLM is primarily determined by the embedding dimension (d_model), the number of attention heads (n_heads), the number of layers (n_layers), and the vocabulary size (vocab_size). For example, the GPT-3 model has 175 billion parameters, with d_model = 12288, n_heads = 96, n_layers = 96, and vocab_size = 50257.

Model Parallelism: Training and deploying such massive models require substantial computational resources and specialized hardware. To overcome this challenge, model parallelism techniques have been employed, where the model is split across multiple GPUs or TPUs, with each device responsible for a portion of the computations.

Mixture-of-Experts: Another approach to scaling LLMs is the mixture-of-experts (MoE) architecture, which combines multiple expert models, each specializing in a specific subset of the data or task. The Mixtral 8x7B model is an example of an MoE model that leverages the Mistral 7B as its base model, achieving superior performance while maintaining computational efficiency.

Inference and Text Generation

One of the primary use cases of decoder-based LLMs is text generation, where the model generates coherent and natural-sounding text based on a given prompt or context.

Autoregressive Decoding: During inference, decoder-based LLMs generate text in an autoregressive manner, predicting one token at a time based on the previously generated tokens and the input prompt. This process continues until a predetermined stopping criterion is met, such as reaching a maximum sequence length or generating an end-of-sequence token.

Sampling Strategies: To generate diverse and realistic text, various sampling strategies can be employed, such as top-k sampling, top-p sampling (also known as nucleus sampling), or temperature scaling. These techniques control the trade-off between diversity and coherence of the generated text by adjusting the probability distribution over the vocabulary.

Prompt Engineering: The quality and specificity of the input prompt can significantly impact the generated text. Prompt engineering, the art of crafting effective prompts, has emerged as a crucial aspect of leveraging LLMs for various tasks, enabling users to guide the model’s generation process and achieve desired outputs.

Human-in-the-Loop Decoding: To further improve the quality and coherence of generated text, techniques like Reinforcement Learning from Human Feedback (RLHF) have been employed. In this approach, human raters provide feedback on the model’s generated text, which is then used to fine-tune the model, effectively aligning it with human preferences and improving its outputs.

Advancements and Future Directions

The field of decoder-based LLMs is rapidly evolving, with new research and breakthroughs continuously pushing the boundaries of what these models can achieve. Here are some notable advancements and potential future directions:

Efficient Transformer Variants: While sparse attention and sliding window attention have made significant strides in improving the efficiency of decoder-based LLMs, researchers are actively exploring alternative transformer architectures and attention mechanisms to further reduce computational requirements while maintaining or improving performance.

Multimodal LLMs: Extending the capabilities of LLMs beyond text, multimodal models aim to integrate multiple modalities, such as images, audio, or video, into a single unified framework. This opens up exciting possibilities for applications like image captioning, visual question answering, and multimedia content generation.

Controllable Generation: Enabling fine-grained control over the generated text is a challenging but important direction for LLMs. Techniques like controlled text generation and prompt tuning aim to provide users with more granular control over various attributes of the generated text, such as style, tone, or specific content requirements.

Conclusion

Decoder-based LLMs have emerged as a transformative force in the field of natural language processing, pushing the boundaries of what is possible with language generation and understanding. From their humble beginnings as a simplified variant of the transformer architecture, these models have evolved into highly sophisticated and powerful systems, leveraging cutting-edge techniques and architectural innovations.

As we continue to explore and advance decoder-based LLMs, we can expect to witness even more remarkable achievements in language-related tasks, as well as the integration of these models into a wide range of applications and domains. However, it is crucial to address the ethical considerations, interpretability challenges, and potential biases that may arise from the widespread deployment of these powerful models.

By staying at the forefront of research, fostering open collaboration, and maintaining a strong commitment to responsible AI development, we can unlock the full potential of decoder-based LLMs while ensuring they are developed and utilized in a safe, ethical, and beneficial manner for society.

0 notes

jcmarchi · 2 days

Text

Dorik Review: The Best AI Website Builder Using a Prompt?

New Post has been published on https://thedigitalinsider.com/dorik-review-the-best-ai-website-builder-using-a-prompt/

Dorik Review: The Best AI Website Builder Using a Prompt?

Building a website can be daunting for many individuals, particularly if they lack graphic design or development skills. And with so many website builders nowadays, how do you know which one to choose?

I recently came across Dorik, an AI website builder that uses artificial intelligence to generate an entire website in seconds based on a single text prompt. With Dorik, anyone can create a professional-looking website without design or coding knowledge.

As someone who has extensively used popular website-building platforms like WordPress and Webflow, I was intrigued by Dorik’s user-friendly approach. Could Dorik AI build an entirely customizable website with a single text prompt? Is it the best AI website builder? I had to try it for myself!

In this Dorik review, I’ll discuss what it is, who it’s best for, and its features. From there, I’ll show you how I made this professional website in seconds: https://unite.dorik.io/

I’ll finish the article by recommending the best Dorik alternatives I’ve tried so you know which AI website builder best suits your needs. By the end, you’ll understand what Dorik is all about and whether or not it’s the right choice for your website-building needs.

Verdict

The Dorik website builder is among the best AI website builders, easily generating customizable websites from simple text prompts in seconds. Despite some limitations like template variety and more robust e-commerce capabilities, its intuitive interface and reliable customer support make it a top choice for effortless website creation. Dorik particularly benefits individuals and small businesses seeking a hassle-free web design solution!

Pros

Generate a customizable website in seconds for free with a text prompt.

The free version has sufficient features for creating a basic website.

80+ templates, 250+ UI blocks, and 20+ elements to create a unique website.

Websites are automatically responsive on any device.

Websites can be built in any language.

The drag-and-drop interface and AI tools simplify customization.

The interface is user-friendly and easy to navigate.

Helpful resources and support like a Helpdesk, 24/7 Live Chat, and Facebook Community.

Easily import and export HTML/CSS/JS code to transfer your website to other platforms.

A free version plus single payment plans with no monthly fees.

Cons

The websites may lack originality, making this less ideal for those seeking custom solutions.

E-commerce capabilities are possible but limited, making it less suitable for businesses seeking advanced e-commerce functionality.

The third-party integration process may be challenging for some users.

The free version has some restrictions, such as limited storage, bandwidth, and using a dorik.io subdomain.

A 200 blog post limit on the personal website plan.

What is Dorik?

With 100,000+ happy customers, Dorik is a CMS, blogging, and website builder that lets you create professional websites with a prompt without the need for coding or design experience. Its technology has been trained on a vast dataset of existing sites, allowing it to understand the principles of web development and design. It recognizes patterns and trends in website layouts and elements, enabling it to generate visually appealing designs tailored to your needs.

To use Dorik to generate a custom website in minutes, all you have to do is describe your website through a text prompt. The prompt is best written like a command and can include information about the website’s purpose, target audience, and desired features.

Using its advanced algorithms and machine learning techniques, Dorik analyzes the text using website-building principles and design best practices to create a visually appealing and functional website! It generates layouts, compelling copy, pixel-perfect images, and customizable UI blocks that match your description.

You’ll also be able to add essential website features and pages, such as contact forms, blogs, and product listings. From there, customize your website’s functionality, SEO settings, and design elements through the intuitive dashboard!

Combining AI technology with web design, Dorik eliminates the need for coding or design skills so that anyone can create a website in minutes with a single text prompt! Dorik takes the technical aspects of website building so you can focus on more important things like content creation and branding.

Who is Dorik Best For?

Dorik is an excellent tool for anyone lacking coding and design skills and is interested in using AI to generate a stunning website in minutes. However, there are certain types of people Dorik is best suited for:

Entrepreneurs and small businesses can use a text prompt to create a professional, personalized website without developers or designers. Plus, with options for adding contact forms, blogs, product listings, customizable UI blocks, and SEO settings, entrepreneurs can focus on business growth and customer engagement instead of technical website development.

E-commerce businesses can use Dorik’s customizable e-commerce templates to create attractive online stores. They provide a free SSL certificate for secure transactions, unlimited bandwidth for handling high website traffic, and seamless integrations with Shopify and Ecwid.

Bloggers can use Dorik’s user-friendly CMS to create and publish content effortlessly. Customize templates to match personal branding, use SEO tools for better visibility, and integrate with social media platforms for seamless sharing and reader engagement. With customizable UI blocks and SEO settings, bloggers can focus on creating quality content while Dorik manages design and functionality.

Web designers can more quickly create websites using Dorik without coding or extensive design experience. With 80+ customizable templates and UI blocks, crafting unique client designs is easy. Designers can collaborate with clients by inviting them to projects and assigning custom roles. Dorik also offers a white-label CMS for agencies to brand and provide the platform to clients.

Dorik Features

Here are Dorik’s features so you know what it’s capable of:

Dorik AI

White-Label CMS

Website Builder

Blogging Platform

Membership Sites

SEO Optimization

Integrations

250+ UI Blocks

80+ Templates

1. Dorik AI

Dorik AI is the technology behind Dorik that generates websites in seconds! Just provide the website name and a short description, and Dorik AI will analyze the text prompts and create a website layout with copy and images in any language.

Dorik AI saves significant time and money than manually designing a website with code or outsourcing this to designers and developers! The websites Dorik generates are stunning, the copy it writes is compelling, and the images are pixel-perfect for any device.

The platform also makes customization easy for users without coding or design experience. Drag and drop elements and easily adjust layouts while maintaining responsiveness to create a personalized website.

Once you’re happy with your website, use GPT-4 to enhance the tone of your copy to resonate the most with your target audience. Cancel any stock photo subscriptions and regenerate images and illustrations with Dall-E 3 directly in the editor! You can even regenerate entire layouts and sections.

From there, publish your website directly to the Internet with the Dorik.io subdomain or a custom domain! It’s incredible how Dorik AI simplifies the website-building process to create professional websites without needing specialized skills.

2. White-Label CMS

For agencies and web designers, Dorik offers a white-label CMS option to brand the platform as their own and deliver it to their clients. It allows you and your clients to generate pages, manage team members, and publish content without needing plugins and additional hosting.

Dorik white-label CMS has everything you need that you and your clients will love:

User-friendly interface for clients to navigate and manage their websites.

Showcase your branding and logo within the platform to reinforce your brand identity.

Complete control over the CMS with client billing, access control, and content management.

Client access to their website’s dedicated dashboard.

Monthly or annual subscription payment options to bill your clients.

The white-label CMS option Dorik provides empowers agencies and web designers to offer clients a fully branded website-building experience.

3. Website Builder

With 40+ integrations available, Dorik offers a comprehensive website builder that simplifies creating a professional website. Rather than manually writing out code, you can drag-and-drop components, buttons, 250+ UI blocks, or 80+ customizable templates to build a website in seconds! No maintenance or plugin updates are required; you’ll get unlimited bandwidth and storage.

With Dorik, you can add images, image carousels, subscription forms, payment buttons, and blogs. To help you design, invite team members to help! Dorik gives you complete control by allowing you to add custom roles and set permissions.

The Dorik AI website builder comes with everything you need:

Reliable hosting and CDN.

A free SSL certificate.

Receive payment through your website via Stripe, PayPal, and Gumroad.

Unlimited website storage and bandwidth.

User analytics to see how visitors behave on your website.

Password protection on any page of your website.

SEO optimization to maximize your reach.

Connect a custom domain from any provider.

Export clean, un-minified HTML/CSS/JS code with media files.

Connect and display Airtable data.

Automatic responsiveness on multiple devices.

Plugin-free blogging.

And more!

Dorik’s website builder means you can create visually appealing and functional websites without coding or design skills.

4. Blogging Platform

Dorik’s built-in CMS includes a user-friendly blogging platform for creating, scheduling, and publishing blog posts. It offers SEO optimization tools, meaning you can optimize your blog posts for search engines to attract more customers, clients, etc. You can also integrate your blog with social media platforms to easily share your blog posts and engage with your audience!

Dorik’s blogging platform is valuable for content creators looking to establish a solid online presence and engage with others.

5. Membership Sites

Dorik makes it easy to create membership sites and offer exclusive content and services to subscribers. This functionality means that you can restrict access to certain parts of your website and provide exclusive content to subscribed members. Integrate Stripe, create payment plans, and add paywalls!

Dorik’s membership site features include subscription management tools, allowing you to handle member registrations, payments, and renewals. Plus, all of your earnings are yours! Dorik doesn’t take a cut from any commission you make!

6. SEO Optimization

Dorik has all the necessary SEO tools to make Google love you and increase traffic to your website. That means more customers, clients, and revenue to your business!

Here are the SEO features that come with Dorik:

Meta Titles and Description

Custom Schema

No Follow-No index

Sitemap.xml

Href Lang

Robots Txt

Whether you want to optimize pages or posts, Dorik has what you need. You’ll easily be able to add meta titles, descriptions, and canonical URLs without writing any code. If you don’t want a page to appear in search engines, turn on “No Index” with a single click!

Site speed and responsiveness are other critical factors in bringing in the most traffic possible. Dorik recognizes this by using a Global Content Delivery Network to enhance page loading speed globally and improve your visitor’s experience and domain rating. Websites built on Dorik are also automatically responsive, meaning they’ll look great on any device!

7. Integrations

Dorik offers seamless integration with many popular third-party services, enhancing the functionality and capabilities of your website. These third-party services extend the functionality of your website to include features such as e-commerce capabilities, analytics tools, and email marketing services.

Integrating with third-party services also eliminates switching between different platforms or manually transferring data, streamlining your workflow and enhancing efficiency. Integrations also provide additional customization options, allowing you to tailor your website to your needs and preferences.

8. 250+ UI Blocks

Dorik offers 250+ UI blocks to build beautiful websites. These UI blocks are in 20+ categories, including SaaS, Web/Mobile Apps, and Personal & Agency Websites.

UI blocks are great because they guarantee design consistency throughout your website. You can easily customize UI blocks to fit your branding and content. Adjust the layout, add or remove elements, and change the styling to create a unique and personalized design.

Lastly, UI blocks save time by eliminating the need to design every section of your website from scratch. You can add professionally designed sections to your website with a few clicks!

UI blocks allow you to easily create a visually appealing website, even without design experience.

9. 80+ Templates

If you want the fastest website creation, choose from Dorik’s library of 80+ templates! The templates are stunning and touch multiple industries, including SaaS, Web/Mobile Apps, and Personal & Agency Websites.

Templates are excellent resources for creating a visually appealing starting point for your website. They are professionally designed and tailored to specific niches for a cohesive and attractive aesthetic. Website templates save time and effort customizing your website to suit your needs.

But you’re not limited to the template’s design! While they provide an excellent starting point, you can easily customize templates to reflect your branding and content. Dorik makes it easy to adjust colors, fonts, and layouts for a unique and personalized design.

Dorik’s templates offer a convenient and efficient way to create a professional website without starting from scratch.

Top 3 Tips for Choosing the Right Template

Here are my top tips on how you can choose the Dorik template that’s right for you:

Consider the purpose and industry of your website. Dorik offers a variety of templates tailored to different niches, such as fitness, food, marketing, and personal portfolios. Choose a template that aligns with your website’s niche!

Pay attention to the layout and structure of the template. Look for a design that lets you showcase your content effectively and provides the most seamless user experience.

Consider the level of customization available with the template. Look for flexible templates that allow you to customize elements to fit your branding and content easily.

How to Use Dorik to Build a Website

Here’s how I used Dorik to generate, customize, and publish a website with a text prompt in seconds:

Create an Account

Create a New Site with AI

Write a Prompt

Generate Website

Customize Website

Publish Website

Step 1: Create an Account

I started by going to the Dorik homepage and selecting “Try it Free.”

Step 2: Create a New Site with AI

After making my account, Dorik took me to the dashboard. I selected “Create New Site with AI.”

Step 3: Write a Prompt

Generating a website with Dorik was incredibly easy, requiring only three pieces of information:

The name of my website.

A description of the business.

The language.

Rather than giving Dorik a basic business description, I made the website description more like a command:

“Create a website discussing everything there is to know about artificial intelligence, including news, interviews, and reviews on the best AI tools. Design it like a news website.”

It doesn’t have to be anything crazy. Keep it simple and concise, and tell Dorik exactly what kind of website you want it to generate.

Once I filled in the information, I hit “Generate Your Website.”

Step 4: Generate Website

Immediately, Dorik started generating a website based on the command I gave it. A few seconds later, I had a fully customizable and responsive homepage! Here is the website Dorik generated for me if you’re curious: https://unite.dorik.io/

Based on my prompt, Dorik did an excellent job generating a homepage; everything looked professional and modern with a great layout. However, I wanted to see what Dorik’s customization tools were like.

Step 5: Customize Website

Once Dorik generated my website, the AI Quick Style panel opened. From this panel, I updated the colors and font of my entire website with a single click.

The rest of the tools were easily accessible on the left:

Add Elements

Components

Pages

Global Style

Navigation

Media Library

Site Settings

AI CMS and Blog

These tools made it easy to design something that fit the branding. I could easily drag and drop UI blocks onto the website to add text blocks, image galleries, and social media icons. The UI blocks are automatically formatted to be responsive on any device, which saves a lot of time rearranging elements!

The built-in AI features also sped up the customization process significantly. I could click on any image or piece of content and rewrite it instantly using AI or generate entire web pages.

Hovering over the website containers gave me even more options. On the top left of each container, I could click and drag to change the container size or move it to where I wanted. I could also edit the container, delete it, or add a new one.

In the middle were my element options. I could move, edit, delete, or add a new element to the container. My favorite was the AI text options to improve, rephrase, simplify, etc., the text with AI. I could even select photos from the website and replace them with AI-generated images!

Step 6: Publish Website

Once I was happy with everything, I headed to the top right of the website editor to preview, save, or publish my website to the Internet!

I hit “Publish,” and Dorik gave me two options for how I wanted to publish my website:

Dorik.io Subdomain

Custom Domain

I didn’t have a custom domain, so I went with the Dorik.io subdomain and hit “Publish Site.”

Overall, I was blown away by how simple Dorik made designing a fast, responsive website in seconds. As someone who has built websites on various platforms, Dorik has been the easiest and quickest!

Top 3 Dorik Alternatives

To ensure Dorik is the right AI website builder for you, here are the best Dorik alternatives I’ve tried that you’ll want to consider.

Divi AI

Divi AI is a popular Dorik alternative specific to the WordPress Divi theme. Like Dorik, it uses AI to generate entire web pages with a text prompt. Within the web builder, it can design layouts, write content, and create images.

Both platforms offer a wide range of customization options and a responsive design without the need for coding knowledge. Choose one of the pre-designed templates to get started quickly, and drag and drop elements exactly where you want them for the perfect website layout.

As you can see, Divi AI and Dorik are very similar. Use AI to generate entire web pages, write content, and generate images.

However, Divi comes with 200+ elements and 2,000+ premade designs. Meanwhile, Dorik offers only 80+ templates, 250+ UI blocks, and 20+ elements to create your website.

If you want to generate a website using AI on WordPress, I’d highly recommend getting the Divi theme and Divi AI. It’s the most popular WordPress theme with many designs and elements to create the perfect website! If you want a better user experience without dealing with plugins, Dorik is the best choice.

CodeDesign

CodeDesign is another AI website builder that approaches web design similarly to Dorik. Give CodeDesign a prompt, choose the language, and it will generate a responsive website in seconds! The content will be SEO-optimized, and customizing the website is easy with its drag-and-drop functionality.

The designs CodeDesign generates look great and are professional. However, if you don’t like what it’s generated, you can regenerate with the click of a button. Alternatively, you can always choose from the 300+ templates CodeDesign offers.

CodeDesign and Dorik are very similar AI website builders offering many of the same features. The only difference that stood out was that CodeDesign provides more templates and better pricing than Dorik.

Regardless, CodeDesign and Dorik are excellent AI website builders. Both platforms offer free plans where you can generate and customize a website for free without a credit card. I’d highly suggest trying both and seeing which suits your needs best!

Hostinger Website Builder

The Hostinger Website Builder is another popular Dorik alternative requiring no coding skills. It has 150 templates, but you must purchase hosting before using the tool.

Generating a website with Hostinger is basically the same process as Dorik:

Describe the website you want to build.

Select the kind of website.

Provide a brand name.

Hostinger will start generating your website, which will take only a minute.

Customizing the AI website is incredibly easy, and Hostinger has built-in AI tools to streamline the process further. Some other AI tools unique to Hostinger include an AI Logo Maker, an AI Writer, and an AI Heatmap!

For more AI tools, templates, and the cheapest pricing, choose Hostinger. If you don’t want to pay upfront, choose Dorik!

Dorik Review: The Best AI Website Builder Using a Prompt?

After using the Dorik website builder to generate and customize a website in seconds, I can confidently say that Dorik is among the best AI website builders on the market. Its AI-driven approach offers a seamless website creation process and transforms text prompts into fully functional websites! With Dorik, anyone can create a fully functional website without design or coding knowledge.

However, while Dorik could have more templates and elements for customization, its user-friendly features, responsiveness, and SEO optimization tools make Dorik a versatile website builder. Meanwhile, Divi AI, CodeDesign, and Hostinger Website Builder are the best options if you’re considering alternatives.

Thanks for reading my Dorik review! I hope you found it helpful. Dorik has a free plan that lets you generate, edit, and publish a website, so why not try it for yourself? Explore Dorik’s innovative solutions to streamline your website creation process efficiently and effectively.

Frequently Asked Questions

How does Dorik work?

Dorik is a website builder that uses AI (Artificial Intelligence) to simplify website creation. With Dorik, you can generate a website from scratch by providing a text prompt. The AI tools within Dorik generate a layout, customize the site’s text and images, and even create AI-generated photos that you can edit. Once you’re satisfied, publish to have your website go live!

Is Dorik free or paid?

Dorik offers both free and paid options. The free version of Dorik has certain restrictions, such as limited storage and bandwidth and using a dorik.io subdomain. Paid plans are available with additional features and functionality, with a one-time payment for lifetime pricing! That means no monthly fees.

Who is the CEO of Dorik?

The CEO of Dorik is Mizanur Rahman. Mizanur and Anwar have been creating websites and templates since 2012, and Dorik was initiated as a side project in 2019. Feel free to read more about their story on the Dorik website.

Can AI build me a website for free?

Yes, AI can build you a website for free in seconds. With Dorik, you must create an account, describe the website you’d like to generate, and hit “Publish.”

However, there are some limitations to the free plan, such as limited storage and bandwidth and the use of a dorik.io subdomain. Limitations will likely be the case for any website that generates free websites using AI. Upgrade to a paid plan to access advanced functionality and customization.

How do I contact Dorik?

To contact Dorik, go to the Dorik Helpdesk and scroll to the bottom to chat or email the Dorik support team. Dorik offers 24/7 chat support with knowledgeable representatives who can assist with inquiries or issues.

What is the best website builder out there?

Dorik is among the best website builders out there. It’s incredibly user-friendly and only requires a text prompt to generate a stunning website in seconds.

The best alternative website builders include Divi AI, CodeDesign, and Hostinger Website Builder. Evaluate your requirements and compare the features and pricing of different options to determine the best fit for your needs.

0 notes

jcmarchi · 2 days

Text

Exclusive Discount on DanteAV PTZ Cameras - Videoguys

New Post has been published on https://thedigitalinsider.com/exclusive-discount-on-danteav-ptz-cameras-videoguys/

Exclusive Discount on DanteAV PTZ Cameras - Videoguys

Save $500 on PTZOptics Link 4K PTZ Cameras

Your Dante Workflow is About to Get Easier Ready to add video to your Dante setup? With Dante AV-H™ and the latest in auto-tracking capabilities, the Link 4K fits easily into even the most complex Dante design or workflow. The Link 4K combines broadcast quality with Dante’s simple discovery, set up and management, simplifying and professionalizing any video production installation.

What is Dante AV-H? Dante AV-H brings device interoperability and the power of Dante to H.26x endpoints. With Dante AV-H, all video and audio routing, configuration, and management of devices is done with one easy to use platform. The Link 4K will automatically sync up with your Dante system clock leader ensuring seamless audio and video synchronization, and is able to output Dante AV-H video on your network natively, along with audio sources connected to the 3.5mm audio input. Designed with the Future in Mind Every Link 4K camera features SDI, HDMI, USB, and IP output for unparalleled performance and versatility. The Link 4K features 4K at 60fps over HDMI, USB, and IP. Using technology by Dante, the Link 4K can also provide 4K at 60fps resolution over Dante AV-H. The Link 4K offers built-in auto-tracking capabilities — no need to run software on another computer — freeing teams of the need for a camera operator.

PTZOptics also makes a line of PTZ Cameras with NDI|HX

Now with NDI|HX license included!

PTZ and auto-tracking features, now more accessible. Combining HDMI, SDI, USB, and IP outputs into one camera, with native NDI® support coming soon. Perfect for live streaming and video production, offering high-quality 1080p60fps resolution & excellent low-light performance thanks to SONY CMOS sensors.

starting at $999.00

Featuring auto-tracking for a more intelligent video production workflow. The Move 4K is capable of 4K at 60fps (1080p at 60fps over SDI), future-proofing your technology investment while still accommodating HD and Full HD video resolutions equipment. The Move 4K offers high performance in low-light, PoE+ capabilities, & a built-in tally light.

starting at $1,799.00

0 notes

jcmarchi · 2 days

Text

The Friday Roundup - Sound Design & Free A.I. Music

New Post has been published on https://thedigitalinsider.com/the-friday-roundup-sound-design-free-a-i-music/

The Friday Roundup - Sound Design & Free A.I. Music

This website uses cookies so that we can provide you with the best user experience possible. Cookie information is stored in your browser and performs functions such as recognising you when you return to our website and helping our team to understand which sections of the website you find most interesting and useful.

#audio #browser #cookies #course #crash #DaVinci Resolve #Design #DIY Video Editor Blog #Music #sound #sound design #tutorial #user experience #Video

0 notes

jcmarchi · 3 days

Text

Stellar Starter Tips For Stellar Blade

New Post has been published on https://thedigitalinsider.com/stellar-starter-tips-for-stellar-blade/

Stellar Starter Tips For Stellar Blade

Stellar Blade is heading into full release, and you can dive into the action-heavy sci-fi adventure for yourself. You can read our full review for more details about how it turned out, or read our interview with the game’s creative director to learn more about how the game came together.

Developer Shift Up does a good job of gradually introducing new core systems as you move through the lengthy adventure, but as we played ahead of public launch, our team discovered several things the game doesn’t tell you that can lead to a more enjoyable journey.

Turn On Auto Loot

While the developer made many smart decisions in creating Stellar Blade, we’re a little confused why the default choice for gathering dropped loot requires pressing and holding one of the shoulder buttons.

As soon as you start the game, pause and go to Settings > Gameplay and adjust the Auto Loot option to “On,” which is not the default. Eve will begin to pick up items and currency automatically without the need for your input – as long as you get close enough to the items on the ground, they’ll go into your inventory.

Set Hair Length

In that same menu we just pointed you toward (Settings > Gameplay) you’ll find another option to adjust Ponytail Length.

This is primarily a cosmetic choice for how you like the main character to look, but we mention it here for functional reasons. Depending on your playstyle, extraneous visual “noise” onscreen can be distracting, especially in a tense encounter.

Eve’s extremely long ponytail shows off some cool physics, but if it’s preventing you from having a clear view of enemy actions, consider changing the ponytail length to “Short”

Ambush When You Can

Even standard wandering Naytiba enemies in Stellar Blade can lead to some tough and even game-ending battles. Take every advantage you can, including prioritizing an early unlock of the Ambush ability (findable in the Attack section of the Skill Tree).

Ambush lets you backstab enemies if you can get close enough before they turn around and see you. It’s generally an instakill, so it’s a good choice for eliminating the most powerful foe in a group before facing off against the less dangerous ones.

Rush To Open

Similar to the Ambush option above, it’s smart to unlock and use the Rush skill early on and use it to open fights where Ambush isn’t an option. This skill lets you close with an enemy instantly, and you can get some hits in before they have a chance to fight back.

It’s also an essential skill in more protracted combats, including bosses. After dodging or running out of the way of a dangerous attack, Rush lets you get back into the fray right away.

Guard First, Then Strike Back

More than many action games, guarding is an essential skill that you should practice from the very start. Excessive offensive play without guarding is a recipe for failure.

Your guard/shield will diminish with hits taken, but it’s an excellent way to confront the early moments of a fight, when you’re still feeling out an enemy. And while perfect parries and dodges are great, you may need a few strikes to recognize an enemy’s attack patterns, and guarding is the best way to weather those moments, with the exception of unblockable attacks.

Improve Perfect Dodge And Parry

There’s no more surefire way to succeed in Stellar Blade than getting good at Perfect Dodge and Perfect Parry. It’s absolutely core to the combat system, and success with those skills opens up additional options and abilities.

Of course, the most important thing you can do here is observe and improve your timing in nailing these abilities. Much of that is about memorizing the different colored flashes and knowing which color corresponds to the need to dodge or parry. The training modules inside the Skills menu can help a lot with that practice. To find those, go to the Skill Tree Survival section, and tap Square while hovering over the Perfect Dodge or Perfect Parry ability. Seriously, there are no better skills to master.

But you can also hedge your bets. There are additional abilities in the skill tree and equipment in the game that make it easier to nail those tight windows for perfect dodges and parries. If you’re having trouble consistently hitting that timing, these are excellent skills and equipment items to focus on.

Scan, scan, scan

A tap of the central touchpad on your DualSense controller will send out a pulse scan of your surrounding area. It’s an easy thing to forget about as you run around slicing through enemies.

That would be a mistake. Making a point to scan frequently is a smart move. It reveals enemy locations that may be hiding. And it also shows critical interactable items and individuals, including dead bodies that can sometimes hold precious upgrades, like max health improvements.

It costs nothing to tap that scan, so get in the habit of doing it in each new area you arrive in.

Keep up with your sidequests

If you plan to try and mainline the golden path of Stellar Blade, feel free to ignore this tip. But for the rest of the players, you may be wondering what to prioritize in your extracurricular activities.

Sidequests provided by actual individuals in the game world are almost always a good bet. They provide more and better experience and rewards than just roaming around taking out monsters. Those character-provided sidequests also tend to be more interesting than the less personal assignments you can pick up from bulletin boards.

And, while we don’t want to spoil anything, it’s safe to say that the latter hours of the game turn out differently depending on certain sidequests you might have chosen to complete. Talking with and completing those character-connected side quests is a great place to put your time if you want to expand your time in the world of Stellar Blade.

Pick up the phone!

It goes without saying that you should enter and use supply camps when you find them in order to rest and get your health and supplies back. But early on, you might not realize that you should also be trying to activate all the payphone locations that you encounter.

Phones activate a new fast travel point; you want as many of those as possible.

Even if you don’t initially plan to do a lot of fast traveling, later missions sometimes send you back to explore areas you’ve already visited, and having a more comprehensive selection of fast travel points gets you back to the action and the quest completions faster.

Unlock Beta and Burst Wisely

As you play through Stellar Blade, you’ll gradually unlock access to the Beta and Burst skills. Each skill tree adds new abilities that you can bring to bear in a fight by holding down L1 or R1 along with one of the face buttons. These are usually high-powered attacks; you should integrate them into your combat flow.

But even knowing that, it’s easy to cast too broad a net and weaken yourself by playing with too many different skills. That’s especially true in the midgame, while there’s still much to unlock.

In both the Beta and Burst skill trees, consider unlocking and completing the upgrades for one (or maybe two) of these skills before moving on to the others. Remember, both Beta and Burst skills pull from their own respective meters, so if you use one of the abilities, it will deplete that meter, preventing you from doing another deployment. It’s rare that you’ll be able to spam out multiple abilities that tap into the same meter. As such, it’s best to focus on individual skills that shore up weakpoints in your combat repertoire.

For instance, Shield Breaker is the bottom skill (L1+X) on the Beta Skill Tree, and as the name implies, it does a great job of breaking enemy shields. If you’re struggling with the early phases of boss fights when your foe often still has a high shield, this is a great choice. And rather than diversifying into other skills, we recommend unlocking the upgrades to Shieldbreaker before focusing on other skills. Alternately, Triplet (L1+Square) is a solid and consistent damage dealer and a great choice if you’re just looking to bring down standard enemies quickly.

Take the time to look through the full suite of Beta and Burst skills, and consider which individual abilities would be the biggest help to your playstyle, then focus on those.

Follow the Green Dot

As you get started in Stellar Blade, you may encounter many individuals you feel you need to talk to. If that’s your thing, go for it. But if you’d like to focus on the individuals that actually have something to offer, look for the green dot next to an individual; these are actual quest-givers or other meaningful conversations.

Take the Hint

Occasionally, you’re going to encounter panels that demand that you enter a code sequence of symbols. Generally speaking, you need to find a different location that provides that code before you can input it.

Don’t bother trying to jot down the symbols. Instead, when you return to the appropriate unlockable panel, tap the “Hint” button to see the code. It’s not cheating. It’s simply showing you the code you already found elsewhere in the game, which matches this particular panel. If the “Hint” option is grayed out, you just don’t have the code yet.

Experiment With Story Mode

We always advise that the best difficulty setting is the one that leads to the most fun for you as a player.

With that said, Stellar Blade features a tight and rewarding combat system – if you can enjoy the challenge of some tough combat, the default Normal Difficulty is the way to go. It has some hard fights, but with patience, you’ll succeed.

With that said, there’s nothing to say you can’t experiment with Story Mode, and doing so can help you prepare for winning a hard fight in Normal (or the eventual Hard mode, which unlocks after you beat the game).

In Story Mode, combat prompts pop up and time slows down in critical moments of battle, letting you learn exactly what buttons to press when that vicious boss descends on you.

If you’re committed to beating the boss on Normal, but you’re having a hard time, consider bumping the difficulty down to Story Mode to learn the ropes of the fight. Then, before you defeat the boss, switch back to Normal mode, which will reset you to your most recent save point. You can then go into the fight armed with a clear understanding of what buttons to press to react to each attack – but you’ll still have to nail the timing to win.

Good luck in your adventures with Eve across the post-apocalyptic wastelands of Stellar Blade!

0 notes