The DeepSeek R1 training procedure confused me at first. My brain refused to accept this powerful model could be incredibly straightforward.

Let me break down this elegant beast for you 🧵 This multi-stage training loop is unusually effective:
Base → RL → Finetune → RL → Finetune → RL

Does scaling stages = better performance? Let’s break down each phase. 🔍

I did some research on LLM as agents today. Here is a guide to the state-of-the-art of LLMs as agents!

It's all about environments where LLMs can observe, plan, act, and iterate on solutions.

🧵1/8 2/5 There are two main benchmarks that are useful. Both are seemingly hard datasets.

Especially SWE-Bench. @cognition_labs was able to show-case 13.86% accuracy with Devin.

SWE-Bench: GitHub issues or pull requests.
MINT: arxiv.org/abs/2310.06770
arxiv.org/abs/2309.10691