Summary of In-context Learning May Not Elicit Trustworthy Reasoning: A-not-b Errors in Pretrained Language Models, by Pengrui Han et al.
In-Context Learning May Not Elicit Trustworthy Reasoning: A-Not-B Errors in Pretrained Language Models
by Pengrui Han, Peiyang Song, Haofei Yu, Jiaxuan You
First submitted to arxiv on: 23 Sep 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
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 explores the inhibitory control abilities of large language models (LLMs) by designing a text-based multi-choice QA scenario inspired by the A-Not-B error, a phenomenon seen in human infants. The experiment finds that state-of-the-art LLMs like Llama3-8b perform well with in-context learning (ICL), but struggle when the context changes, demonstrating a significant drop in reasoning tasks of up to 83.3%. This suggests that LLMs have inhibitory control abilities comparable to those of human infants, often failing to suppress previously established response patterns during ICL. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper is about how well artificial intelligence (AI) can think and make decisions. It’s like a big test for AI computers called language models. These language models are very good at understanding and answering questions when they’re given specific information. But, just like babies, they have trouble stopping themselves from doing something even if it doesn’t work anymore. In this experiment, the researchers tested how well these AI models could make decisions when the rules changed. They found that the best AI models got stuck in old habits and had a hard time learning new things. |
