Deciphering the Factors Influencing the Efficacy of Chain-of-Thought: Probability, Memorization, and Noisy Reasoning
arxiv.org
Chain-of-Thought (CoT) prompting has been shown to enhance the multi-step reasoning capabilities of Large Language Models (LLMs). However, debates persist about whether LLMs exhibit abstract generalization or rely on shallow heuristics when given CoT prompts. To understand the factors influencing CoT reasoning we provide a detailed case study of the symbolic reasoning task of decoding shift ciphers, where letters are shifted forward some number of steps in the alphabet. We analyze the pattern of results produced by three LLMs -- GPT-4, Claude 3, and Llama 3.1 -- performing this task using CoT prompting. By focusing on a single relatively simple task, we are able to identify three factors that systematically affect CoT performance: the probability of the task's expected output (probability), what the model has implicitly learned during pre-training (memorization), and the number of intermediate operations involved in reasoning (noisy reasoning). We show that these factors can drastically influence task accuracy across all three LLMs; e.g., when tested with GPT-4, varying the output's probability of occurrence shifts accuracy from 26% to 70%. Overall, we conclude that CoT prompting performance reflects both memorization and a probabilistic version of genuine reasoning. Code and data at this https://github.com/aksh555/deciphering_cot
  • Lugh@futurology.todayOPMEnglish
    4·
    6 days ago

    Although I’m using AI more and more for writing related tasks, I still find it constantly making simple rudimentary errors of logic. If it is advancing as this research paper claims, why are we still seeing so many of these types of hallucination errors?

    • MajorHavocEnglish
      2·
      6 days ago

      If it is advancing as this research paper claims, why are we still seeing so many of these types of hallucination errors?

      I mean, the research could be true, while the AI is merely achieving a level of reasoning of a houseplant or a bug, while streaming randomized garbage the rest of the time.

      That would still be a promising sign of progress, while not of any current practical use.

      (Edit: which is guess is pretty often what science progress looks like.)