#Deep tech | Explore Tumblr Posts and Blogs

nokiareptilestonaniteandlumen · 10 months

Text

Dear Readers, How's it going? Good, I hope.

For the last 6 months I’ve been keeping my eyes peeled for anything AI related that I could understand, and today, I’m finally going to weigh in on Sam Altman’s brainchild, ChatGPT.

First disclaimer, I’m not a professor, nor an expert. Nowhere near it and since I started this blog about 1 year ago, I have been pursuing a career in tech, so I’m a noobie. I thought I’d go down the rabbit hole of neural networks because lots of experts are hinting towards it as the future. With a formal education in communications from the Univ of Tenn @Chatt, I know enough about research, media, and business to be dangerous. Second disclaimer, since this new wave of tech was recently released, I have had the pleasure of picking at it abroad, therefore in two different languages (3 if you count Wolfram, a computational language). My research will be presented at the bottom, however, I mainly relied on 2 sources because I simply preferred their direct approaches. Warning - were about to get into the weeds, ***obligatory gulp of coffee***

First was an interview conducted by The New Yorker. In my intro I asked you guys, “How’s it going?” I bet you didn’t answer like ChatGPT, “ As a language model, I do not have the ability to experience or do anything. Is there anything else I can assist you with?” You probably sound more like this, “I’m fine, thanks.” Quite the different approach, but exactly the same as Siri. This is important because it is how we differentiate chat bots from humans. Which leads to the second question - why is it that you are a you then if you aren’t a sentient being? This makes me think of men and their cars. For me, I have only talked to a car to pep talk it into making it to the next gas station. You spend your good and bad times with your car. When you aren't a grease monkey/mechanic and she makes a weird noise or doesn't do what you want, people resort to talking to it and make loosely based comments based on these behaviors sometimes to appease it or treat it. We could go further with the similarities of these relationships, but the analogy screeches to a halt because cars don’t talk back, unless you're Chuck Norris.

Back to the interview, the answer it gives is interesting albeit creepy. It says it’s for you (the user) to feel more natural. Evidently, our brains aren’t wired to speak to AI. But it’s this inauspicious start that sets the mood for the article and makes the AI seem unsettling at the least and perhaps a little manipulative.

My research then went to the tech side from watching Stephan Wolfram do a 1-hour breakdown on his blog that I think is worth checking out. If you’re in a hurry, I have taken my time to bring you my highlights. His perspective is one of greatness, as a CEO of an eponymous research company, and a neural network researcher.

I wanted to learn how the technology works and be able to explain it in broad terms before testing or adapting to everyday life. Like in life, it’s always best to gain knowledge of something foreign, before blindly collaborating with or passing away precious past time with it. This topic was different than most. It was hard to read about on platforms like Twitter. These sites thrive off of outrage, I was coming to this conclusion after laboring through posts that only boiled down to shock value. Or as the writer, Bounthavy Suvilay (Indie Games 2) aptly puts it, (they) ‘only benefit social media networks by keeping their users captive in a heightened emotional state’. I’ll add to this, they are a great place to find pessimism as compelling as it is obscure.

So what is it? ChatGPT is based on the fact that there is regularity in the English language, and it may be even deeper than we thought, it takes this structure (grammar, literary tools,etc.) and assimilates what we know. As you know its goal is to complete your text, but it does this by taking everything it’s dealing with and grounds it up to numbers called weights as opposed to a computer which operates in 1s and 0s.

After this, the AI uses what it knows about the English language and returns (at a rate of 1 word at a time) the outcome and that’s as far as I can understand technically. Again I’m not a computer scientist so I’ll stop there and leave you with the quote, “the simplest answer is usually most likely the correct one”. What is it with ChatGPT’s super celebrity status though, why are so many people becoming users? Its wild success in the short time it’s been available makes apps like Instagram seem novel. I don’t get it. But I was obsessesd with the movie Phenomenon featuring the John Travolta. Is it the ol saying if it’s free it’s for me… Most of the internet world can speak or understand the English language. This might be a helpful start.

Back to Wolfram, in the Q&A portion of his blog, I loved how he wistfully entices his audience by flaunting his 45 years of expertise casually stoking the fire of the deeptech industry, which has been around for years. Experts consider 2012 a milestone when Googlex found it possible to train and use deep neural nets. Concentrating on ChatGPT, it's not only scraping the internet, it’s picking up regularity in the way humans speak/write similar to how we learn. But some aspects may be deeper and it’s likely picking up haptics from a space where we have yet to be able to artificially describe. Maybe that last part is a stretch and unprovable, but may be as the tech inevitably progresses. In the end, Wolfram draws parallels with other aspects of biology and says in theory these features can be attributed to other animals. He was vague but sounds a little like Dr. Doolittle to me.

On this animal topic, let’s take a dog, any kind, your family pet, a sheep dog, or even a police dog. According to Meta’s chief AI scientist Yann LeCun, ChatGPT in its current evolutionary stage resembles a canine, and it will take the next 30 years to reach human intelligence.

But back to my question, what makes this app different? My take is when people seek new toys/games/etc., especially ones that try to fool the brain, we get this stubborn and relentless urge to test its limits until one is fulfilled. And in this respect, ChatGPT has passed with flying colors. If you have tried the app, take the example of ‘tokens’. OpenAI engineers are like “the house” in a casino except instead of cards they deal in workability, the game is how closely can their ‘tokens’ work to sound like logically sounding answers according to human’s current understanding of the topic. This token can be reinserted into this neural net until one’s tiny heart is desired, at the same time the next prompt is fed back into the machine working in its favor as feedback. Until, you can no longer trace the token back to its original form — meaning you cannot ever truly arrive at a perfect answer. The boundaries are also limited by how many tokens can be used. And to reduce server usage, OpenAI started limiting tokens.

I don't want to mince words, but they haven’t sold me on it. I decided to learn about it before trying, and I’m glad I did. They essentially released the beta to collect data, but that’s not why I turn up my nose. It’s my background in sales, I have to be sold on stuff before buying it/using it. And frankly, the world obsessing over something is not enough to interest me. The pessimist in me still strives to find utility. For now I’ll stick with Google. I know it’s different and old skool, but in the end they use algorithms that take your words, or what ChatGPT refers to as prompts and quickly lead you to an answer that still satisfies my little heart.

I really loved the spirit of how creatives saw the utility in strong-arming sucky machines with it. I’m referring to this Foxbusiness.com article where it tricked a task rabbit by playing a person who is blind in order to forgo a CAPTCHA. Sounds like a wee-bit Black Mirror, duuuude. I had to investigate further on the subject to find out visually impaired are truly struggling with CAPTCHA. Something I never thought about. I then uncovered some even cooler news. This minority who has trouble seeing can now use ChatGPT to ID things in photos. Side note: what a terrible security system CAPTCHA is. I’d argue this invention is as annoying as the pop-up.

Also, I want to address people profiting from AI-written books by selling them via sites like Amazon. I doubt these guys are actually making money, if so awesome, but as someone who reads I don’t buy it. From a Reddit thread on the other hand, I learned that video game devs are using the LLM to write code. However, it is uber specific code in the video game engine Unity. In fact, it helped code blades of grass to appear more realistic. You can’t just write into the prompt code grass moving and basta! The coder is already skilled and delegates tasks to the AI to save time.

In the end we will undoubtedly come up short in fixing all of society’s problems via using it in its current form, and like most tech advancements, they will likely aid in generating wealth for Big Tech. Speaking of, Reddit is now being hijacked by its most popular mods and (***puts on tinfoil hat***) to my belief, it might have something to do with pressure created from companies like OpenAI's. Why? The threatening of ad revenue perhaps, why sift through hundreds of Reddit comments threads when the machine does it for you. More specifically ChatGPT's operation depends on ‘terabytes of books and Reddit posts, virtually all of Wikipedia and Twitter, and other vast repositories of words’, according to The New Yorker, or as Wolfram estimates ‘a trillion-ish words of texts’ are at its disposal.

Speaking of disposal, let’s not get started on its environmental impact. As I painstakingly try to sort my trash from recycling, ChatGPT servers are sitting in an air-conditioned warehouse 'plagiarizing (sic) essays, sending flowery emails and asking if God exists,' says Aisling Ní ChúláinNo’s (euronews.com article).

At last, we all know when it comes to freemium software or ones being sold for a loss, it’s only the tip of the iceberg. ChatGPT at first seemed to me like a beefed up predictive text finding the most plausible of ways to explain ideas via language, but now I know its use is gaining potential and has a 30-year plan to take the world by storm. I’d like to push it further and interview an OpenAI employee next month.

**RELATED FUTURE BLOGPOST lol ** - The new wave of enthusiasm for neural networks created by the release of ChatGPT appears promising for the future of big tech with its eco-friendly rating being harmful for its stakeholders.

SOURCES:

What is ChatGPT doing...and why does it work?

https://www.newyorker.com/news/the-new-yorker-interview/its-not-possible-for-me-to-feel-or-be-creepy-an-interview-with-chatgpt

Suvilay, Bounthavy. Indie Games 2. Portland, Oregon, Ablaze LLC, August 16, 2022

#tech #chatgpt #twitter #neural networks #deep tech #biology #french #franglais #futurism #innovation #meta

3 notes · View notes