Self-Discover: Large Language Models Self-Compose Reasoning Structures

Abstract: 

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.

Background Knowledge:

Understanding of Frontier LLM Reasoning Research

Bio: 

Steven Zheng is currently a Senior Staff Software Engineer and Tech Lead at Google DeepMind working on frontier LLMs. He tech-led and made significant contributions to multiple critical LLM projects at Google, including LaMDA, PaLM-2, Bard, Gemini 1.0 and Gemini 1.5. Steven has extensive research and model development experiences from Pretraining, Post-Training to Inference-time techniques. His work has been highlighted by Google executives and has received several awards including Google Tech Impact Award (H1 2023). Steven received his PhD in Quantum Physics and Quantum Computation from Duke University, focusing on quantum waveguide-QED, quantum phase transitions and quantum computations. He has published 50+ papers in journals and conferences such as Nature, Nature Physics, Physical Review Letters, ICLR, ICML and NeurIPS etc. Recently, his research has been focusing on benchmarking and improving SoTA LLM’s core Planning and Reasoning capabilities.

Open Data Science

 

 

 

Open Data Science
One Broadway
Cambridge, MA 02142
info@odsc.com

Privacy Settings
We use cookies to enhance your experience while using our website. If you are using our Services via a browser you can restrict, block or remove cookies through your web browser settings. We also use content and scripts from third parties that may use tracking technologies. You can selectively provide your consent below to allow such third party embeds. For complete information about the cookies we use, data we collect and how we process them, please check our Privacy Policy
Youtube
Consent to display content from - Youtube
Vimeo
Consent to display content from - Vimeo
Google Maps
Consent to display content from - Google