Denny Zhou
Denny Zhou is a research scientist at Meta Superintelligence Lab, and his research centers on prompting language models to reason in steps. He joined Google Brain in 2017 and founded its reasoning research group. He co-authored the chain-of-thought prompting paper, published in 2022, and the self-consistency and least-to-most prompting papers, published in 2023. He was also on the PaLM team behind the 540-billion parameter Pathways language model, and in 2024 he co-wrote a paper on how chain of thought lets transformers solve inherently serial problems.
Timeline
-
2017
Joining Google Brain
In 2017 he joins Google Brain and founds its reasoning research group.
-
April 2022
Member of the PaLM team
In April 2022 Google Research describes PaLM, a 540-billion parameter Pathways language model, and thanks the PaLM team, which includes him.
-
2022
Chain-of-thought prompting
He co-authors the chain-of-thought prompting paper, published at NeurIPS 2022, which shows that intermediate reasoning steps significantly improve complex reasoning in large language models. In May 2022 he and Jason Wei post a Google Research blog about the method.
-
2023
Self-consistency decoding
He co-authors a self-consistency paper at ICLR 2023, which proposes a new decoding strategy to replace the naive greedy decoding used in chain-of-thought prompting.
-
2023
Least-to-most prompting
He co-authors a least-to-most prompting paper at ICLR 2023, which breaks a complex problem into simpler subproblems and solves them in sequence.
-
2024
Serial problems and chain of thought
He co-authors a paper at ICLR 2024 with Zhiyuan Li, Hong Liu and Tengyu Ma, showing how chain of thought lets transformers solve inherently serial problems.
-
Since 2026
Research scientist at Meta
He joins Meta in 2026 as a research scientist at Meta Superintelligence Lab.
Sources and credits
Photo: We have no published photo of Denny Zhou yet. Send us a free-licensed photo.
- Google DeepMind's Coding Pivot Lost Six Researchers to Meta, OpenAI, and Anthropic
- Pathways Language Model (PaLM): Scaling to 540 Billion Parameters for Breakthrough Performance
- Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
- Language Models Perform Reasoning via Chain of Thought
- Self-Consistency Improves Chain of Thought Reasoning in Language Models
- Least-to-Most Prompting Enables Complex Reasoning in Large Language Models | ML Anthology
- Chain of Thought Empowers Transformers to Solve Inherently Serial Problems