Summary of Preemptive Answer “attacks” on Chain-of-thought Reasoning, by Rongwu Xu et al.

Main: Computation and Language (cs.CL)
Secondary: Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR)

Preemptive Answer “Attacks” on Chain-of-Thought Reasoning

by Rongwu Xu, Zehan Qi, Wei Xu

First submitted to arxiv on: 31 May 2024

GrooveSquid.com Paper Summaries

GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!

Summary difficulty	Written by	Summary
High	Paper authors	High Difficulty Summary Read the original abstract here
Medium	GrooveSquid.com (original content)	Medium Difficulty Summary The paper investigates the robustness of Large Language Models (LLMs) coupled with Chain-of-Thought (CoT) prompting against preemptive answers, which can impair the model’s reasoning capability. The authors introduce a novel scenario where the LLM obtains an answer before engaging in reasoning, and show that this significantly hinders the model’s performance across various CoT methods and datasets. To address this issue, they propose two measures aimed at mitigating the problem.
Low	GrooveSquid.com (original content)	Low Difficulty Summary The paper studies how to make Large Language Models (LLMs) better at thinking through problems. It finds that when an LLM gets an answer too quickly, it can’t think as well afterwards. The authors call this a “preemptive answer” and show that it happens even with the best prompt methods. They also suggest ways to fix this problem so that LLMs can keep thinking clearly.