research For AGI, we need better tasks. For better tasks, we need open-endedness. (ALOE 2022 Notes) Open-endedness is the idea of a system that "endlessly creates increasingly interesting designs". That's a simple concept with big implications. Biological life on earth is diverse and beautiful and the result of open-ended evolution. Human culture is open-ended; we've created a lot of art
thoughts Kvfrans Thoughts: A Mathematical Definition of Interestingness A Mathematical Definition of InterestingnessIn open-endedness research, we care about building systems that continously generate interesting designs. So what makes something interesting? Previously, I argued that interestestness depends on a viewer's perspective. But this answer isn't very satisfying. Ideally, we would want some mathematically-definable
research To extract information from language models, optimize for causal response One way to view AI progress is that we can measure increasingly abstract concepts. A decade ago, it was really hard to take a photo and measure “how much does this image look like a dog?”. Nowadays, just pick your favorite image classification model
thoughts Kvfrans Thoughts: Oct 29 This post is a bunch of random thoughts about things I've read recently. It's organized into sections, and the sections don't really relate to each other, but here they are. Neural Networks are Data DigestersA common philosophy around my head is that "representation learning
research CLIPDraw: Exploring Text-to-Drawing Synthesis CLIPDraw synthesizes novel drawings from text. Play with the codebase yourself. AI-assissted art has always been intriguing to me. To me, visual art is a very human thing -- I can’t imagine a way that a computer rediscovers our own cultural concepts without
research StampCA: Growing Emoji with Conditional Neural Cellular Automata StampCA growing emoji. Play with the codebase yourself. When a baby is born, it doesn’t just appear out of nowhere -- it starts as a single cell. This seed cell contains all the information needed to replicate and grow into a full adult.
research Quality Diversity: Evolving Ocean Creatures If you've ever folded a paper airplane, chances are you folded The Dart. It's fast and simple, and kids all around the world have launched it through the air. Soon, however, they will move on to other designs – airplanes that glide, make loops, or
research Open-Endedness 3: Multicell World How many shapes can be made from six 2x4 LEGO bricks? The answer is 915,103,765. By combining only a few simple parts, it’s surprisingly easy to achieve high levels of complexity. A Lego Counting Problem: http://web.math.ku.dk/~eilers/
research Open-Endedness 2: Bitwise Chemicals When algae is in the pond, a fish will come to eat it. When fish are in the pond, birds will come to catch them. This is an example of peer dependency: a species’ survival depends greatly on the presence of other species. Peer
research Open-Endedness 1: Cellworld When I look into the pond at my local park, there’s an amazing ecosystem of life at play. Algae covers the surface, and small fish come up to nibble on the meal. In the mud, bottom-feeding clams and large eels wait for prey.
games Omakase 5: Bullet Dance This month was a return to form. Ever since Rain Project, I've always had this urge to make a true boss battler, combining platforming and bullet hell. With November's Omakase, I finally took the dive with Bullet Dance. I'm super proud of this month's
games Omakase 4: Jopper For October's game I collabed with the great Jynnie Tang!! Which basically means we came up with an idea late night over juice and decided "hey let's try and make it happen". The core concept we wanted to reach was the feeling
games Omakase 3: Diplomacy "Diplomacy" is a rather infamous game among board game enthusiasts -- It's an experience known to destroy the deepest friendships and create mistrust that can brew for years. That was a bit of exaggeration, but there's a key reason that Diplomacy has
games Omakase 2: Suwapond With this month's Omakase game, I wanted to take a super simple game idea, and working on polishing the feel of the core mechanic -- not necessarily the mechanic itself, but how satisfying it is to the player. I was also traveling for a
link [Link] How we built the Waifu Vending Machine We taught a world-class artificial intelligence how to draw anime. All the drawings on the left were made by a non-human artist! Wild, right? It turns out machines love waifus almost as much as humans do. We proudly present the next chapter of human
games Omakase 1: Ropeman Every month, I'm going to build a game. Each one will be small, and in each game I'm going to focus deeply on a single aspect and try and improve myself at it. At the end, I'll release the game along with a writeup
Linking C++ and Python with Boost (Anaconda) So this just a quick post, detailing a fix that I spent a good half day trying to get find. I'm currently working an idea for cool RL algorithms involving a bunch of agents living in an interactable world (think dwarf fortress, rimworld etc.
research [Link] Automatic Sketching and other Vectorizations! With Canvas-Drawer networks. During my summer at Autodesk, I was tasked with building a computer agent to sketch out images. Here's how.
games RAIN Project: evolution of the game development dream Eleven months ago on a long train ride home, I wrote the first lines of code for a small platforming game. Little did I know that this prototype was the start of something much more than a just game -- it was a dream
research [Link] Learning a Hierarchy How can agents learn hierarchies that are useful across a wide range of tasks? At OpenAI, I set out to answer this question.
research Deepcolor: automatic coloring and shading of manga-style lineart In the past few months, I've been experimenting with digital art. The process involves creating a rough sketch, cleaning it up into lineart, and then finally coloring and shading to create a final product. Out of these three steps, adding color is by far
tutorials What is the natural gradient, and how does it work? A few months ago I attempted to understand the natural gradient, and wrote a post to help organize what I knew. Unfortunately there was too little detail and all I really understood was a "black box" version of the natural gradient: what
research Speeding up TRPO through parallelization and parameter adaptation If you look through the results on the OpenAI gym, you'll notice an algorithm that consistently performs well over a wide variety of tasks: Trust Region Policy Optimization, or TRPO for short. However, the algorithm does take quite a bit of time to converge
research Colorizing the DRAW Model In my last post I described the DRAW model of recurrent auto-encoders. As far as I've seen, the only implementations of DRAW floating around Github deal with the MNIST dataset. While they are helpful for reference, I wanted to have a model that could
tutorials What is DRAW (Deep Recurrent Attentive Writer)? A few weeks ago I made a post on variational autoencoders, and how they can be applied to image generation. In this post, we'll be taking a look at DRAW: a model based off of the VAE that generates images using a sequence of