We introduce SELF-DISCOVER, a general framework for LLMs to self-discover the
task-intrinsic reasoning structures to tackle complex reasoning problems that
are challenging for typical prompting methods. Core to the framework is a
self-discovery process where LLMs select multiple atomic reasoning modules such
as critical thinking and step-by-step thinking, and compose them into an
explicit reasoning structure for LLMs to follow during decoding. SELF-DISCOVER
substantially improves GPT-4 and PaLM 2's performance on challenging reasoning
benchmarks such as BigBench-Hard, grounded agent reasoning, and MATH, by as
much as 32% compared to Chain of Thought (CoT). Furthermore, SELF-DISCOVER
outperforms inference-intensive methods such as CoT-Self-Consistency by more
than 20%, while requiring 10-40x fewer inference compute. Finally, we show that
the self-discovered reasoning structures are universally applicable across
model families: from PaLM 2-L to GPT-4, and from GPT-4 to Llama2, and share
commonalities with human reasoning patterns.

SELF-DISCOVER 是一个通用的框架，使 LLMs 能够自我发现任务内在的推理结构来解决复杂的推理问题，并在 BigBench-Hard、基于代理的推理和数学等具有挑战性的推理基准上，相较于 Chain of Thought (CoT) 的性能提升高达 32%。此外，SELF-DISCOVER 在需要 10-40 倍少得多的推理计算的情况下，超过了 CoT-Self-Consistency 等推理密集型方法 20% 以上。最后，我们证明了自我发现的推理结构在模型族之间是普适的：从 PaLM 2-L 到 GPT-4，从 GPT-4 到 Llama2，并与人类推理模式共享共同点。