Why We Think
中文摘要
John Schulman provided feedback, improving model performance through test-time compute. Recent research explores its effective use.
English Summary
This post examines recent advancements in test-time compute and Chain-of-thought reasoning, investigating how increasing "thinking time" enhances model performance and why these techniques are effective.
Special thanks to John Schulman for a lot of super valuable feedback and direct edits on this post. Test time compute (Graves et al. 2016, Ling, et al. 2017, Cobbe et al. 2021) and Chain-of-thought (CoT) (Wei et al. 2022, Nye et al. 2021), have led to significant improvements in model performance, while raising many research questions. This post aims to review recent developments in how to effectively use test-time compute (i.e. “thinking time”) and why it helps.