Summary: Today, llms are trained in a multi step process post SFT. RL -> Generate quality synthetic data → Self-Distillation on that → another round of RL (simplified). Fable-5 had a solution strategy constrained on how to compose code, and…
Some problems don’t get solved in your head, they get solved in someone else’s while you watch or read their solution. You might feel it before you understand it. A small calibration. A quiet “Oh!” You cannot always tell what got them there…
Some elements here build upon the RL env post. Also, please read the footnotes, given a lot to cover, I moved many clarifications to the bottom. Something is happening across all major labs simultaneously and it’s not a coincidence. When ev…
I occasionally like to reverse engineer computer games by playing against inbuilt AI and trying different tactics to find loopholes. One such game is Catan Universe. Playable in a browser and built in Unity and WebGL. I wanted to see if an
AI trends move fast. Writing about them makes sense if it’s your job. I’m picking this up because it’s escalated recently, and some discussion seems needed. This is me trying to make sense of an RL environment (and this understanding may ev…
Richard Sutton wrote about the ‘Verification Principle’ in 2001: An AI system can create and maintain knowledge only to the extent that it can verify that knowledge itself. I like the principle stated in just the previous line better though…
someday i will make a more thorough and better reasoned post around it but at this point these are just notes somewhat structured The current discourse around LLM memory feels fundamentally misguided. We’re stuck in this paradigm of “how do…
From Scott Alexander’s 2015 post Douglas Adams once said there was a theory that if anyone ever understood the Universe, it would disappear and be replaced by something even more incomprehensible. He added that there was another theory that…
In 2019, Richard Sutton, wrote his groundbreaking essay titled ‘The Bitter Lesson’. Simply put, the essay concludes that systems which get better with higher compute beat the systems that do not. Or specifically in AI: raw computing power c…
Ballon D’Or, awarded by France Football to the best footballer in the world, used to be a very respected award. There were some disputes but it was by and large agreed that whoever got the award was deserving of being called the best in the…
AI is everywhere and the space is moving so fast, no one knows where all this is headed. Seeing this unfold, here is an idea of small bets. You may know a few things about research, or perhaps you have intuitions, let’s have some fun guessi…
Talking to a few investors about raising for Clio AI, I had this sudden realization: currently most (if not all) AI investments are correlated. Investment funds are operating under the following assumption: This is a great thesis with a cou…