Summary of Failure Modes Of Llms For Causal Reasoning on Narratives, by Khurram Yamin et al.
Failure Modes of LLMs for Causal Reasoning on Narratives
by Khurram Yamin, Shantanu Gupta, Gaurav R. Ghosal, Zachary C. Lipton, Bryan Wilder
First submitted to arxiv on: 31 Oct 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Computation and Language (cs.CL)
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 Large language models (LLMs) struggle with causal reasoning, relying on shortcuts like topological ordering or parametric knowledge. Despite being state-of-the-art, they fail when events are not presented in exact causal order, and long-term causal relationships are difficult to infer. LLMs also rely too heavily on their pre-existing knowledge, degrading performance when narratives contradict it. Synthetic experiments and real-world evaluations show that explicitly generating a causal graph improves performance, while chain-of-thought approaches are ineffective. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models (LLMs) have trouble understanding cause-and-effect relationships. They make mistakes by relying on shortcuts instead of thinking deeply about the situation. For example, they might assume that earlier events cause later ones just because of the order in which things happen. This can lead to problems when the events aren’t presented in a way that makes sense. LLMs also have trouble understanding long-term cause-and-effect relationships and may not be able to figure out what’s causing something if it takes many steps. The good news is that by creating a map of the causes, these models can get better at making predictions. |
