Gh: github.com/karpathy/autoresearch Really cool project that both shows the trend for agents and what the core loop for them is, although graphs can still be deceiving Always love a Karpathy post!

Andrej Karpathy just put out a GitHub repo for an AI agent that does research for you. And if you don't know who he is, he's one of the biggest names in AI. He was one of the founding researchers at OpenAI and the senior director at Tesla. But this project is really cool because it both highlights the main trend and shows how things work very clearly. As Andrej Karpathy himself said, he thinks this will be the decade of agents, where the past few years we've seen LLMs explode in productivity, but now the focus will be building things around it, like Cloud Code or OpenCloud. And fundamentally, most of them follow a pattern of three steps. Do a thing, evaluate how you did the thing, and improve how you did the thing. And that framework is really useful for seeing how AI can impact industries and knowing what things it can and cannot be used for. And second is that things aren't always what they're promised to be. If you look at this graph, it looks really impressive on first glance, but if you look at the y-axis, it goes from 1 to .975. I don't know what this example was for, but I don't think a 2.5% increase is usually very groundbreaking, even though the concept is pretty impressive. So this is both a great example of how agents work and what they can be applied to, and also an example to be skeptical of how powerful these agents are. So I wanted to share that with you all. Cheers.