Hendrycks, D., Burns, C., Kadavath, S., Arora, A., Basart, S., Tang, E., Song, D., and Steinhardt, J. (2021). Measuring mathematical problem solving with the math dataset. https://arxiv.org/abs/2103.03874  

Large Language Models are Zero-Shot Reasoners https://arxiv.org/pdf/2205.11916 

Solving Quantitative Reasoning Problems with Language Models https://arxiv.org/abs/2206.14858  

Chain-of-Thought Prompting Elicits Reasoning in Large Language Models https://arxiv.org/abs/2201.11903 

Self-Consistency Improves Chain of Thought Reasoning in Language Models https://arxiv.org/abs/2203.11171 

 Large Language Models for Mathematical Reasoning: Progresses and Challenges https://arxiv.org/abs/2402.00157

Tree of Thoughts (ToT): Deliberate Problem Solving with Large Language Models https://arxiv.org/abs/2305.10601 

Buffer of Thoughts (BoT): Thought-Augmented Reasoning with Large Language Models https://arxiv.org/pdf/2406.04271  (Given a specific problem, they retrieve a relevant thought-template and adaptively instantiate it with specific reasoning structures to conduct efficient reasoning)

TO COT OR NOT TO COT? CHAIN-OF-THOUGHT HELPS MAINLY ON MATH AND SYMBOLIC REASONING https://arxiv.org/pdf/2409.12183 (Sept. 2024) Main result: CoT gives strong performance benefits primarily on tasks involving math or logic, with much smaller gains on other types of tasks

Better Zero-Shot Reasoning with Role-Play Prompting: https://arxiv.org/html/2308.07702v2 (can apply role play with zero shot in class)

Faithful Logical Reasoning via Symbolic Chain-of-Thought https://arxiv.org/abs/2405.18357 

MAMMOTH: BUILDING MATH GENERALIST MODELS THROUGH HYBRID INSTRUCTION TUNING https://arxiv.org/pdf/2309.05653 (uses hybrid Chain of Thought, and Program of Thoughts to achieve higher accuracy on MATH Data set.

MATHVISTA: EVALUATING MATHEMATICAL REASONING OF FOUNDATION MODELS IN VISUAL CONTEXTS https://arxiv.org/pdf/2310.02255 A new data set, MATHVISTA, to benchmark on mathematical and visual tasks.  This paper also discusses the emergent "self-verification" capability of GPT-4V, where GPT-4V is able to verify that some answers are not possible due to contraints of a problem.

CHAIN-OF-VERIFICATION REDUCES HALLUCINATION IN LARGE LANGUAGE MODELS https://arxiv.org/pdf/2309.11495 Chain-of-Verification (COVE) steps are as follows.  For a given question, model first (i) drafts an initial response; then (ii) plans verification questions to fact-check its draft; (iii) answers those questions independently so the answers are not biased by other responses; and (iv) generates its final verified response.