@ts_floydStanford CS329A: Self-Improving AI Agents Fall 2025 just got a 1-hour guest lecture from Anthropic and Google engineers. They walk through self-learning AI systems, then go inside an LLM to show self-correction and backtracking. The agent loop gets broken down from goal to loops to graphs to feedback, plus 6 agent orchestration patterns. They also dig into the generator-verifier gap and why self-improvement stalls. It’s all about planning, multi-step reasoning, and self-correction. Worth the 70 minutes, and there’s an article to go with it. #Anthropic
ดูโพสต์ต้นฉบับ
