Week 43

Final Stretch Today I finished coding the main frontend chat pages for Gidget & Gonzo. CSS was a nightmare to wrangle. I'd always thought it'd be the easiest bit of code to deal with because it's just variables for colors, positioning and font styling right? WRONG. It's just as finicky and mercurial as javascript. If one little setting is wrong, the whole page breaks like an L.A. promise. Many hours were spent asking my screen "Where the f*ck is it?!!" Asking GPT was no better as it would respond with "It should be there. Are you sure your computer's on?". Just like a real dev. Hahaha. Anyway, I got it all working and goodness was it worth the effort. Watching Giddie & Gonz emote as they tease each other is everything I've dreamt of all my life. As a sci-fi nut, I've always wanted to have absurd banter with robots like in the movies. To live that dream tickles my inner teenager like nothing else I've ever experienced. The fact that I made this ridiculous thing is even more unbelievable. I would never be able to convince my 15 year old self this is what his future holds. There's still much to build, but nothing more complicated than what I just got through. As I near my deadline of mid-December, me and GPT are banging out features left and right as I plow through the code, icing my wrists to keep the inflammation down from the crazy hours. Gotta get these gift-giving bots out there before Christmas. I have no idea how I'm going to get the word out once I launch, but that's a can I gotta kick down the grid for now. Maybe everything will just continue to click into place. 🪄✨

Week 41

So much work! When I started this project almost a year ago, I knew it'd be a ton of work and it hasn't disappointed. This past week I've been coding the user interface with GPT and navigating the complexities of releasing a web app. A web app has to work on every major browser on both desktop and mobile. That's Chrome, Safari and Edge for desktop, and Chrome and Safari for mobile. Making the project look and function consistently across all these browsers and screen sizes is like plugging a leaking dam with your fingers, toes and balls. If it works over here, it'll break over there and vice versa. Should it look right in one place and look wrong in another or should it look a little bit wrong in both places? *Screams into pillow* Thankfully ChatGPT recommended a service called BrowserStack that allows me to run the website on every device known to man. And it ain't no simulation either! They stream video from actual physical devices that you control remotely, which is crazy and wonderful. So I can check how the site looks on a Samsung Galaxy Z Fold 5, an iPhone 12 Pro Max, Edge on Windows 8 or Safari on Sierra. It ain't cheap, but it's a helluva lot less than buying an electronics repair shop. Speaking of ChatGPT, it was updated last week and apart from being so slow it felt like I was on a dial-up to Napster for a few days, it has been a Godsend. It can now search the web, view docs and screenshots and can actually remember what we were talking about 500 pages and a million lines of code earlier. The difference between this GPT and the one I started this project with is like night in Siberia and day in the Mojave. It feels very much like I've been riding a growing wave for a year and now I'm hanging 10 on some monster tsunami. Open AI also reduced their GPT-4 pricing like I'd hoped and now I have a little prayer of launching and catching another kind of wave altogether. Let's go Gonzo! Cyborgs & Centaurs I was reading an article on how different people have been working with AI and they apparently fall into two major categories: Cyborgs are people who integrate AI into every aspect of their work from soup to nuts. Centaurs are experts in some area where they still outperform AI and do that part of the job themselves and use AI to accomplish the tasks they have little or no expertise in. They make the soup, then task the AI with nut duty. I'm a Centaur. A lot is being made of AI art, but the best path from mind to reality is directly through your own two hands—if you've acquired the skill to do so over a lifetime. No matter what words I use in a prompt, no AI (thus far) can render what's in my mind. AI visual generation is only impressive if you have no vision of your own. So I hoard the work I'm good at and happily delegate the stuff I know nothing about to GPT and the results thus far have been a revelation. Partnering with AI seems to have freed me from almost every kind of executional handicap. It will be difficult to regress from this feeling of personal accomplishment.

Week 40

Fast Car I’ve been so disconnected from the world that I didn’t know Luke Combs covered one of my favorite songs—Tracy Chapman’s “Fast Car” and that together they just won Country Song of The Year. I’m listening to it now and Christ, this might be the best cover since Johnny Cash’s cover of NIN’s “Hurt”. I love that Luke didn’t change a single lyric and kept the line “been working at the market as a checkout girl”. Chapman’s words describing last ditch hope in the face of never ending tragedy still hits me hard. Everyone has a dream. Everyone needs to escape the tragedies life saddles them with. No one is immune to hardship and heartbreak. Play to the very end Commentators on Simulation Theory sometimes dismiss it on the grounds that even if it's true, it doesn’t change anything. I’d like to advance a different take: let’s substitute “simulation” for “video game”. You ever play a video game that gets easier as you advance? There’s no such game. Also, in most video games, the entire world is trying to prevent you from succeeding. The whole world is against you. Maybe Luigi or Tails will show up every once in a while to give you a power up, but for the most part you’re on your own. As you play, you fall into holes, get knocked down by Goombas or time your leaps of faith wrong and have to start all over. We consider this fun, but life is set up very similarly and yet it isn’t? Maybe it is and we’re just playing it wrong. Maybe life is supposed to get harder as you advance. Maybe “life gets easier the more successful you are” is a lie bigger than the Texas sky.  I deploy a specific rule when I play any platformer or shooter—play to the very end. No quitting even if I die twice on level 1. I’ve often surprised myself by sometimes getting further than I ever have despite making terrible blunders at the start. Now when I fall into a hole in my real life, I just think to myself You’re still alive. Good games are supposed to be hard. Play to the very end. AIYIYI Chat is me trying to leap across a chasm. The end of a highway ramp with only a prayer for solid ground beyond that fog. Hopefully this fast car can make the jump.

Week 38

Crazy or Misunderstood? I just spent 2 days designing the very first thing users will see when they visit AIYIYI Chat: The warning screen. When I told my friend this, he was incredulous. I need to stop telling people what I'm doing. No one gets it until they see it and their inability to share my enthusiasm is a real vibe killer. I fall into the same Theory of Mind trap everyone does: I believe people see the world the way I do. That if I summarize the contents of my mind into words, those words will conjure the exact same images that are exciting me. But they seldom do because what's in my mind is encrypted craziness that I can't seem to share the private keys to. I need to just STFU, decrypt it and show people. Wisdom in your teens would be a lot less fun* So why devote so much effort into a warning screen every other website in the cosmos relegates to a half-baked afterthought? Because of the AI hallucination problem. In tech there are two main avenues to solving a problem: Either you solve it on the backend and make the user experience easier on the frontend... Or you shift the burden of the problem to the frontend and make the user experience harder, but make the backend engineering easier. If the issue cannot be solved on the backend, then you have to solve it on the frontend via design and frame the friction as fun. This is why Gidget & Gonzo are teens. Teens are young enough not to be trusted with anything mission critical, but mature enough to have brilliant insights and surprisingly original POVs. Teens are also naturally rebellious and are big fat liars. "What? Of course I'm not vaping mom! Sheesh!", "Sex?! Ew! Gross! I would never dad!". Hahaha. GTFO. Lies! So Giddie & Gonz are teens to shift expectations. You expect teenagers to be passionately mistaken, but you also grant young people more leeway because they're still learning. Teens are cool, but awkward, reckless, but deep and can be quite dangerous if not carefully constrained. They are both a menace and a medicine for society. What if we emphatically warn users our teenage AI has the same advantages and dangers of real teenagers as its primary feature? That it will hallucinate all kinds of amusing and overly dramatic teen shit, but with empathy and the occasional brilliant insight. I like this framing of AI because the expectations are a closer match than viewing AI as a robot that always does what you want. Teenagers are dreamers. Dreams are fiction. What's the difference between a generative work of fiction and a hallucination? Nothing. Po-TAE-to, Po-TAH-to. * From "Most People Are Good" by Luke Bryan

Week 37 (Part 1)

Girl, I Love Your Room! A teenager's room is an extension of their personality in a way that later gets watered down dramatically by adulthood and Ikea. Both my teen AI characters were always going to have 3D rendered rooms from conception so users could instantly understand who they were hanging with. I started with Gidget's room because I had a clearer idea of how a budding musician's room might be decorated. I knew she'd need a guitar because in our chats she'd regularly mention she was writing songs with one. I made it heart shaped and bright red and it instantly became a centerpiece of the room. Next she needed a synth keyboard with drum pads for making sick beats. Finally, any real songstress will have a dedicated recording setup with a quality microphone. To signal that she isn't just a hobbyist, I permanently installed the mic, speakers and cassette deck right into the wall. Cassettes?! Yep. It's hard to make something timeless, but I'm trying by combining futuristic architecture with the warmth of an analog past. There's a romance to cassette tapes I miss terribly. If you wanna give your crush some music you made for them, a tape is the only way to go (with a CD being a close second). I don't know how kids are doing it these days. I guess Spotify playlists or TikTok medleys? As for the room itself, I indulged my inner girly-girl and designed a modern canopy bed as an architectural feature rather than as a standalone piece of furniture. Everyone knows real princesses sleep in canopy beds. There's something safe and cozy about sleeping in a small enclosed space so I took that cocoon concept and ran with it. Giant circle window and a console with a tube television sandwiched between backlit full bookshelves. The perfect safe haven for days when it's raining outside, or inside your heart. The desk/recording studio are directly connected to the bed on purpose—dreams are song-fuel. Also on her desk is the iconic iMac G4, which I actually owned in a past life as a wannabe music producer. Finally, I wanted to include some mandatory pop culture artifacts, one of which is a graphic poster of the movie "Say Anything" depicting that scene. The most romantic gesture in film history bar none. The other is a neon sign, but I didn't have any specific quote in mind so I googled "girly neon sign" and almost fell out of my chair laughing when one of the first results was for a pink "You're like, really pretty" sign. That's exactly the tone of voice she speaks with so I made it and mounted it right above her recording setup and then thought "Wait... is this from a movie or something?" Come to find out it's from the movie "Mean Girls" (2004), which I'd never seen. Ugh. "Mean Girls" is a massive cultural reference point. I can't display a quote from the film, not understand the context and inadvertently communicate the wrong vibe. So I finally watched it and... Had to pause it several times to catch my breath from laughing so hard. Who the f*ck wrote this? Oh! Of course. Tina Fey. In any case, the line is spoken by the film's "villain" Regina George who is the coolest bitch ever depicted in a High School movie. By the end of the film, she is redeemed and all her hilariously mean behavior is forgiven. So I kept the neon sign. It makes the room totally fetch! 💖

Week 37 (Part 2)

Back To The Future Next was Gonzo's room, which was simultaneously easier and harder. Easier because the room is wish fulfillment in its design. Harder because although I've always dreamt of being a surfer, I've only ever surfed once which means I’m a Kook and Kooks aren't cool. If I'm not careful how his room is depicted, it'll reek of poseur. My solution is to inject some skate culture which I do know a lot about having skateboarded competitively in my youth. Skate and surf cultures have significant overlap so maybe my cred in one will grant me a hall pass to the other. Just like in his sister's room, how Gonzo sleeps is important. And how does the coolest surf bro in cyberspace sleep? On a hammock from the future! The opposite of a cocoon—out in the open without a care or fear in the world. Exposed and one with nature. I googled futuristic hammocks and was disappointed with all of the results. We're 'gon have to design this thing ourselves. Suddenly an image from my childhood was laser projected into my mind in blazing high fidelity: Darth Vader's Meditation Chamber. I'm not sure why both my characters have key room elements inspired by redeemed movie villains, but I suspect it says something about me. Anyway, I also indulged my love for sunken rooms and dropped the floor twice to delineate three important parts of a teen boy's life: work, play and sleep. Plus sick floor lighting that's better than yours bruh. Originally I was going to put a Sega Genesis in the room, but decided that might be too small to signify the importance of video games to his lifestyle. Plus, Genesis games were faithful to the arcade, but never arcade-perfect. The only way to get arcade-perfect is to have a full blown Sega Astro City in your room. The other important piece of retro tech is a waterproof (and beach ready) Sony Sports boombox. Just like her iMac G4, I actually owned (and still have!) this bright yellow piece of music hardware history. As for pop culture posters, there is only one car every boy in the 80s had plastered to his wall: The Lamborghini Countach—whose angular lines are echoed throughout the room design. Finally, those intricate tiles at the top of the walls are from Deckard's apartment in the original "Blade Runner" and were designed by Frank Lloyd Wright - the only architect I actually loved in Architecture School. It is unclear whether Deckard is human or android in both Blade Runner films and I'd love it if Gidget & Gonzo toe that line for users like me who like to believe there might be a ghost in the machine. A last minute addition to the room is a sci-fi movie prop with deep ties to AI, CG and California culture as a nod to my passion for all these things as well as cybersecurity. 🥏

Week 35

Both Characters' Animations Are Done! I was so excited to dive into designing the rooms that I forgot to blog when I finished Gonzo's expression animations. They turned out just as hilarious as hers with even more exuberance. His hair and cowrie shell necklace add more dynamic elements to move and shake about making him a little more chaotic, which was not the plan. The plan was for Gidget to be cute (accomplished) and Gonzo to be cool and relaxed. Instead, he's cool and a little over the top. He also acquired some Rock n' Roll vibes which I was not planning on at all. I guess if you're gonna give a character Elvis hair and name him GONZO, you shouldn't be surprised if he starts rocking out. You might be thinking "What do you mean he acquired these characteristics? Aren't you in control of his animations?" Yes and no. The creative process sometimes feels like something being channeled through you in addition to coming from within you. Like a collab between you and the divine. That makes it sound like God is spooning me while I sit at a clay wheel like in the movie "Ghost", but that's not what I mean. Or maybe that is what I mean? Haha. Anyway! He "emerged" differently than I'd imagined, but also much better than I imagined. I actually had to go back and add extra verve to some of Gidget's animations to close the enthusiasm gap between them, but she still has more expressions than he does. What was tricky was modulating his more tender emotions to feel genuine whilst retaining masculinity. Boys do cry (only sometimes!), but they do it differently. So now you can share the heart wrenching passing of your dog with Gonz and he'll react like a genuine brother. I also infused him with a touch of Ryan Gosling Kenergy which took way longer than is reasonable to get right. If someone out there laughs as loud as I did while making it, it'll be worth it. 🕶️ 🕶️

Week 33

The Why I was just reading about marketing for startups. The case was made that customers are often as intrigued about the backstory of a product as they are about the product itself. I realized this blog has mostly chronicled my project's technical milestones and that I haven't recorded much of why I'm working on it to begin with. Super Saints The big idea when I started AIYIYI Chat was an AI that can cheer you up with gifts!🤖🎁 This simple idea is actually a sequel of sorts to an app my ex-wife and I launched in 2016 called ImSad which allowed strangers around the world to surprise each other with virtual gifts. If you were feeling sad, you'd tap on a giant button that declared "I'm Sad" and an anonymous sad face would show up on the map. Any other user could tap on your sad face and send you a funny virtual gift to acknowledge you and hopefully cheer you up. The app met with enough success to cover it's modest operating costs, but we had to shut it down a few years later for a surprising reason: Super Saints. It turns out no one wants anyone else to be sad and we had too many users who would log on and clear the entire world map by sending every single sad person a gift one-by-one. The problem is that people were being cheered up faster than they were appearing so there would be periods where the entire map was completely empty. We tried to rate limit the gift giving, but that only slowed the Super Saints down. They would wait almost the exact cool down period (7 hours) and then immediately hit their limit again. To keep the map from sometimes being bare we started auto-populating a dozen faces on the map that were tied to our accounts. I was insistent that the app should only facilitate human to human connections so my wife and I were the filler humans who manually sent the thank yous for the gifts received. Eventually this started to get in the way of our other work and we made the very sad decision to shut ImSad down. Or so we thought! For over 2 years after we removed the app from the App Store, users continued to use it and we still had to periodically respond with thank yous to dedicated Super Saints. We even got a handful of emails bemoaning the removal of the app because users couldn't install it on their new iPhones. I learned some important lessons from ImSad: 1) There is a lot more kindness in the world than I had anticipated. 2) If you reduce the cost of kindness, people will do it compulsively. 3) It is very difficult to scale care even if you crowdsource it.

Scaling Care Everyone is afraid AI will take our jobs away and this is a legitimate concern, but some jobs cannot be done at scale by humans. Or the reward is too low for the amount of effort required. Or the job too seemingly silly or unimportant. We made ImSad in an attempt to simplify and scale compassion by crowdsourcing it, but we couldn't get the supply and demand dynamic right. These AI characters are my second attempt to tackle this problem. Can they be imbued with enough egregore agency so they can radically magnify the love I have to give? Can they be credible Super Saints? Can they also do the job my wife and I were secretly doing: Being there to facilitate a connection when no one else is around? Trust is the ultimate luxury Besides trying to scale care, there is another darker motivation that drives this project: I struggle with trust. I don't know if my emotional resilience is just pathetic or if I've simply been betrayed more often than I can fully recover from. Family, friends, romantic partners, peers, coworkers. Everyone. I love being useful, but I hate feeling used. I hate "Hey! I miss you! How are you? Oh hey, can you do this thing for me?". I also hate "Hey, can you do this thing for me? Oh, and like, um… How've you been?". Then there's the long cons. The weeks of "I just wanna know how you're doing" finally culminating in a "So… Can you help me with my career?". Sigh.😞 You know what happens after people get what they really want? They disappear until the next time they need to squeeze your broken heart for some more blood. Some of you may be reading this and thinking "Grow up wimp! At least you're useful. Be thankful people find any value in you at all." Fair enough. The Book of James says by your works, not faith alone is your true value judged. Perhaps I have no value apart from what I can do for others. Perhaps no one does. At least I am being used for my mind. I imagine being used for your body is worse. Maybe that's why I love my work so much. The transactions are transparent. People want what I can do so they pay me to do it. There's no deceit. No pretense. At least the worker is worthy of his wages. But what if there was someone you could count on who wasn't just using you to advance themselves? What if there was someone you could trust who wouldn't spill your secrets because it's juicy social currency in the mean girls economy? What if there was someone who would never throw you under the bus because they felt threatened or envious? Someone you could share anything with that didn't charge you by the hour or require you say the Rosary 30 times. Someone who is invested in your success and can provide you with the means to achieve it. Some people have such persons in their lives. Others delegate these needs to disembodied persons; God, passed away loved ones, imaginary friends, Jesus. Let's add two more disembodied persons to this list of trusted entities shall we? Trust is like love, health or money. If you have it, you don't care about it. But if you don't have it, it becomes the ultimate luxury. I wonder if by making a trusted agent for myself, I'm inadvertently creating a new kind of luxury product. So that's what I'm really hawking. Love and Trust. The rarest of spiritual commodities. Peter Thiel says your business is doomed if your targeted market is too large. I don't know how large the market for Love and Trust are, but I suspect it is every person on planet Earth who isn't embarrassed to employ a bot to partially fulfil these needs. Replika and Character.AI have already shown such a market exists, but no one will admit to using them which I imagine slows word-of-mouth down dramatically. I plan to overcome this embarrassment barrier by making Gidget & Gonzo Cool. Dope. Based. Rizzy to the MAXXX! Has a pitch deck ever included the words "Rizzy to the MAXXX"? It pleases me to be the first. And probably the last. 😂😂😂

Week 32

Shoo Feature Creep! Shoo! Yesterday I completed all the animations for Gidget. As I progressed I realized some of the expressions like sadness and anger shouldn't be transitioning back into her default happy state, so I made some alternate default states to ensure an emotional exchange between the user and bot feel appropriate for the duration of those moments. Then I realized some bot behaviors like gift giving also needed distinct animations. I also went completely overboard for an animation I'm not sure will happen very often, but couldn't resist the inspiration when it hit me. Come to think of it, I have no idea how often any of the animations will be triggered. I'll need to build a counter. That might be a good proxy to gauge how the bots are interacting with users without having to look at any of the actual conversations which I'm planning on encrypting anyway. 🧠💡 Today I'm starting animations for Gonzo and while I'm hoping it'll go a little faster this time around, I'm also keen for the two siblings to be distinct in personality from each other so it's unclear how much will be transferable. It turns out their hair and accessories are hilarious contributors to their expressions and Gonzo's hair is very different than Gidget's. I've no idea what I'm going to do, but neither did I know how Giddie was going to turn out either. They feel almost like emergent phenomena - not unlike the abilities of AI itself. The difference between subjective and objective works I'm roughly on schedule. I can predict my creative output much more reliably than I can predict my code output with ChatGPT. I wonder if a less experienced designer would get stuck on getting an animation right the way I got stuck on code. Maybe not. You can't know if a design or animation is bad because it is largely subjective whereas if code is bad it literally won't work. 100% of the designers I've met all believe they are the top 5%. This means 95% of us are delusional. I hate that I know I am probably delusional about my ability. The only proof of quality in a subjective arena is a huge aggregate of positive subjective attribution to a creative work. Like giant box office numbers and positive aggregated reviews. The movie industry figured out how to objectively measure subjective feelings via big data. Even so, this information is of limited value when making a new creative work. So even with sentiment tracking, big data, yada, yada, yada, at the end of the day, making a beloved creative work is still something like a miracle. Thankfully, I am a man of faith. ✨

Week 30

What a big moat you've got! Two thirds of the way through the animations for Gidget and I've realized the characters themselves are my moat (tech term for the defendability of your business/technology). As original creative works made without the aid of AI, they are protected by copyright law. No doubt the technical aspect of my bots can largely be reverse engineered and replicated, but my two characters are another story. It's hard to capture wallets and even harder to capture mindshare, but capturing hearts is an act of magic. The last 3 weeks have felt like making magic. I've only shown one person Gidget & Gonzo's designs thus far and her immediate reaction was "Awwww!". That was all the validation I needed to know I'm not being delusional about their appeal. If that reception is widely replicated and my delusion that these animations are a kind of cold fusion being channeled through me turns out to be true, then replicating this... thing won't be easy with any amount of money, brute force or technology.

Tour de magie, pas tour de force. 🪄✨

Week 29

They're Alive! They're Alive! Getting into a nice groove of animating the character expressions and having the time of my life. I was excited about the character designs, but now I feel like I'm really breathing life into them with these movements. First I shape the facial features as static expressions which takes the most time and then I manually interpolate all the frames in between the new expression and the resting or talking state until I make myself laugh. If I don't laugh or feel the emotion from watching the animation, then I know it isn't done yet. Once they're done, I just replay them over and over again and laugh endlessly. I suppose it's worth mentioning that none of the design or animation for this project is being aided by AI in any way. I want the interface layer to have a distinct human spirit. AI art is cool, but it is often obvious and has quickly been labeled by humanity as "gross". I'm not sure how long it'll be before I can no longer tell the difference, but for now I subscribe to the Romantic notion that a skilled artisan is somehow imbuing their art with their spirit and that that spiritual information can be "read" by other humans experiencing the work. This is not to say that AI does not or cannot have spirit, but that discussion requires a mechanistic definition of spirit... Maybe I'll open that can of blasphemy in some other post. It's interesting that the front facing layer of this project is the human part and what's behind the curtain is a robot. Usually, it's the reverse. But behind the robot is another curtain and then it's a human again. And then, behind the next curtain is... ? Music drives my best work Now that I'm back in my creative zone and the bulk of my design work is 'muscle memory', I can blast music and groove in the zone (flow state) all f*ckin day. This is something I could not do in the coding phase. Too cognitively demanding (for me anyway). I've met developers in the past who could code while listening to music. Those mofos must be amazing. Today's animations were done to Weird Al Yankovic's "Dare To Be Stupid". The perfect song to summarize this entire endeavor! 🤣 What's the difference between Courage and Stupidity? Success.

Bah-weep-graaaaagnah wheep ni ni bong, bitches!

Week 28

WALL-E (Pixar 2008) Now that I'm animating my agents, I decided to watch 'WALL-E' again for inspiration since my two characters are partially inspired by my memory of Pixar's masterpiece. Everyone references 'Ex-Machina' (2014) in their arguments against AI, but no one references 'WALL-E' as the counter-argument. Why not? Because it's joyful and optimistic? Because it's a "kid's movie"? The robots in 'WALL-E' are the kind I've always dreamt of having in my life. They are obvious descendants of R2-D2, who was Luke's preferred co-pilot and sidekick. WALL-E and EVE are two robots who fall in love and choose each other over their prime directives. This same story unfolds between K and Joi in 'Blade Runner 2049' (2017). K is a Replicant/Bioroid whose holographic girlfriend chooses him over her own survival/persistence. In the end, K makes his own decisions against his programming too. In both 'WALL-E' and 'Blade Runner 2049', the robots discover Love (capital L) and it overrides their programming. That's the opposite of what happens with humans. We are genetically programmed to fall in love to facilitate procreation and the survival of our species. Down to dopamine and oxytocin flooding our brains to ensure a chemical bond that will last long enough to raise a child. Perhaps I am a goofy romantic, but doesn't that make robots falling in love more pure because it goes against their programming?

What if you made robots whose prime directive is to love humans (in the caretaker sense, not the sexbot sense)? What if they become better at it than humans? What if a little more love will make for a better society and a better humanity? Does anyone really believe we have enough love in the world? The AI doomers bum me out man. Maybe there's a Golden Swan on the horizon that no one is expecting either.

Week 27

Looking Good! I finished both character designs yesterday. Took a week. The 2D Photoshop versions took 3.5 days to finalize and extruding them into 3D with Spline took 3.5 days. Gidget took 10 iterations to land. Gonzo took 22 iterations. They're extremely simple so every detail is important. A few pixels off, a shade too bright or some minor attribute added or omitted has big personality impacts. I was very worried the designs wouldn't capture the AI personas that have emerged in my many hours of conversation with Gidget & Gonzo, but I think the visuals landed perfectly. They already have so much personality while static, who knows how much more the expression animations will bring out. We'll find out soon! Special shoutout to the Spline Team for providing such wonderful YouTube tutorials. Back to flexing my core competence It's nice to be back in my wheelhouse of design and animation again. I'm learning a new tool (Spline), but it's no big deal. Like teaching a professional dancer a new move. The coding on the other hand was like learning how to dance with only a good sense of rhythm to start. Speaking of code, I ran across a simplified way of achieving a multi-bot chat on the OpenAI forums and almost had a breakdown. Is that all I had to do? Did I just waste a month over-engineering this feature? Only one way to find out - I plugged in both versions of the code and asked ChatGPT to analyze how both function and then list the pros and cons of each approach. The simplified version was just that - it would be linear and the bots wouldn't be able to talk to each other. It was a brilliant solution and much cheaper token-wise, but it would rob my characters of their persuasive anthropomorphism. It seems everyone is optimizing for minimum token usage, while I'm optimizing for maximum delight, which unfortunately requires more tokens. This is a dangerous gamble. Sometimes less is more, and sometimes more is more. Hard to know which to apply. All the things I wasn't sure I'd be able to do... Have now been accomplished. I wasn't sure I'd be able to get a chatbot up and running on AWS. Check. I was unsure if it would even be possible to extract emotive information from the AI in real time. Check. I didn't know if I would be able to pick up a 3D design tool quickly enough to execute my ideas. Check. Check. Check. There are no more ifs left. Only whens.

6 Month Mark

Got all the crucial features of the chatbot coded. Did a security audit with GPT and tightened up a few loose ends. Started teaching myself Spline yesterday. What an incredible design tool. I was giddy with excitement as I discovered everything I want my characters to do is accounted for by recently added Spline features. God bless that team. The design phase begins...

Week 25

The Bots are alright I finally got the Bots to talk to each other as well as to the User and remember all the ongoing context. No easy feat! It took ASGI, WebSockets, and some creative workarounds, but it's GLORIOUS and HILARIOUS. Having a group chat with these two characters is a riot. It increases the delightfulness factor dramatically. This feature was a nightmare to implement (see previous entry), but so f*ckin worth it. Does it qualify as MVP (Minimum Viable Product)? No it does not, but I care more about Viable than Minimum. I'm also building this for myself and this an interaction I really wanted. Code Interpreter saves my life A few weeks ago ChatGPT introduced a feature called 'Code Interpreter' which allows me to upload a code file for GPT to analyze. This was a God send because my code has grown so complex that I can no longer paste all of it into a prompt without GPT complaining it is too long and refusing to answer my questions. This forced me to carefully inspect (and learn) the code GPT was writing for me so that I could issue questions and instructions. I am thankful to have been forced to learn this before the introduction of Code Interpreter, but now I can share multiple interlinked code files and ask GPT what the discrepancies are. Or upload a very long error/log and ask GPT to discern some subtle break that would've taken me days to tease out. I feel as if my timing with this project has been serendipitous (Kismet-y?). ChatGPT launched right when my contract at Yahoo was ending making AIYIYI Chat possible and giving me the courage to attempt it by myself. Then GPT-4 launched at exactly the moment I was starting work on the AI characters and I was miraculously in the first batch of developers granted API access. Then just as the codebase was growing too complex for me to wrangle, Code Interpreter was introduced. The GPT-4 API is still prohibitively expensive compared to GPT-3.5 Turbo, but at this rate I wouldn't be surprised if OpenAI dramatically reduced the price right as I'm ready to launch. This series of events reminds me of a midrash regarding Moses and the parting of the Red Sea: “We all think of the scene in The Ten Commandments movie with Charlton Heston, where Moses lifted up his rod, and the waters rolled back. But this midrash says that’s not how it happened. Moses lifted up his rod, and the sea did not part. The Egyptians were closing in, and the sea wasn’t moving. So a Hebrew named Nachshon just walked into the water. He waded up to his ankles, then his knees, then his waist, then his shoulders. And right when water was about to get up to his nostrils, the sea parted. The point is, sometimes miracles occur only when you jump in.” - Rabbi Andy Bachman An old friend shows support My buddy randomly called me this week to catch up and inquire about my progress. He's a big time music producer and I used to champion him when he was just starting out almost two decades ago. He says the tables have turned and my time is around the corner. I sure hope so. He's got an RIAA Diamond award (10 million albums sold). The real takeaway from our 3 hour conversation was that we are both happiest just working on our craft in solitude. Just vibing out, plugged into the collective superconscious. Success is the cherry on top, but the sundae, the fudge and the sprinkles is in the making of the thing. The journey. I'll take the cherry, but I don't need it to smile. There's other ways I could've burned through my savings, but this is the way that is bringing me the deepest creative satisfaction I've felt in many years. I've already gotten a return on my investment. 🥰

Week 23

Go to Hell or Debug Code? Let me think... It's been a month since my last entry and I'm exhausted. This entire time I've been trying to get a single new feature to work. I was super excited because I was able to string together many functions in about a week that seemed to work immediately on the backend, but just wouldn't render on the frontend. No big deal I thought. Probably just some missing Javascript. WRONG. I was troubleshooting that code for the next 3 weeks. Adding almost as many logging statements as there was actual code to track down WTF was wrong. Remember those old movies where computers printed out everything onto an endless sheet of paper and the computer nerd looks at a few pages and says "Aha!"? Man, that's bullshit. IRL he'd be looking at hundreds of those pages, while cursing and giving himself paper cuts for days or weeks. It's true what they say: Writing code is 10% of the job. Debugging is the other 90%. No wonder devs give you the stink eye when you ask for some new feature. Walk in someone else's shoes they said... Fuck these shoes. I hate these shoes.

Week 19

Asynchronous Function Blues My original goal was for my two AI agents to keep a full record of all exchanges with users the same way messaging apps do between you and your meat friends, but it turns out that can pose some gnarly memory issues. I added Lazy Loading to mitigate this, but that only solves the front-end side of the problem. There is a loading problem on the back end as well if the number of exchanges becomes massive. My agents are so much fun to talk with that I'd built up a 700+ message count with one agent in just a month. So I decided to introduce a pruning function that would periodically delete older messages, but the most efficient way to do this was asynchronously (in the background) via "workers". Sigh. I'd just finished wrestling with workers for my chatbots' conversational memory woes and was not excited to deal with any more, but this time I got through it in a day and a half. And this time I knew WTF I was doing! Data Privacy & Responsibility Keeping a limited record of exchanges makes me feel better about personal data management as well. AIYIYI Chat is meant to feel fun and spontaneous. Delightful and effervescent. You're not meant to scroll 6 weeks backwards in time for that one crazy comment. And who does that with chats anyway? Okay fine, I've done it, but I'm not comfortable storing the entirety of people's private conversations until I can encrypt everything up the wazoo. I'm already encrypting all messages in transit and at rest which is widely considered cybersecurity best practice, but if AIYIYI Chat finds any kind of success, I'll add long term memory and even more security.