English Español Français Deutsch
← Back to the list
Reasoning and decomposition Methodologists Directly upstream of ai100's methodology

Kevin Chen-Chuan Chang

Organizing the rapidly growing literature on language-model reasoning into something a researcher new to the area can actually navigate.

By mid-2023 the number of "LLMs can reason" papers exceeded any individual reader's capacity to keep up — chain-of-thought, decomposition, planning, self-consistency, in-context learning, each with multiple variants and competing claims. Chang's "Towards Reasoning in LLMs" survey (2023, with Jie Huang) gave the field its first reasonably complete map of the territory, organizing techniques by mechanism, evaluation paradigm, and empirical findings about what does and doesn't transfer across tasks. Reading the survey is the cleanest way to acquire shared vocabulary with the literature without having to re-derive it from a hundred individual paper abstracts.

Worth following when
you need to onboard quickly to LLM reasoning research and want a structured map that's been peer-reviewed rather than a curated reading list.
Topics
survey methodology for fast-moving subfields; taxonomy of LLM reasoning techniques; the structure of the LLM reasoning literature.
Key works
"Towards Reasoning in Large Language Models: A Survey" (2023, with Huang); long-arc body of work on data mining, deep-web search, and NLP+IR from UIUC.