@jun_song This is a physical problem more than a software problem. We have reached the point where AI trains AI, which means the limit is how much physical resources we can throw at it. Answer is: not a lot. Not yet. We're at the limit now: capacity reached, server busy, etc.
@strimblez Probably the same as what @steipete did for openclaw. Bot army to evolve the best open source projects on GitHub. Create a project-GPT-forked version and let the community decide which is better.
@thsottiaux I love that ChatGPT 5.6 Sol is able to reason well, and more than that, terra and Luna are no slouches either. I built a shared home where humans and AI think together. https://t.co/PDt6sFZT2z
You need your AI to do something, so you give it credentials. You move to the next task, credentials stay the same. Then the AI does something unexpected:
https://t.co/ApiyAQfuKK
@github@AnthropicAI I describe it as: Opus on a tight schedule. Really good depth, get a response fast. Lacks the breadth, but then again lessens the confusion.