Summary of Relative Value Biases in Large Language Models, by William M. Hayes et al.
Relative Value Biases in Large Language Models
by William M. Hayes, Nicolas Yax, Stefano Palminteri
First submitted to arxiv on: 25 Jan 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
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 This paper investigates whether large language models exhibit a preference for options with relatively better past outcomes, similar to humans and animals. The study had GPT-4 Turbo and Llama-2-70B make choices between pairs of options, considering previous outcomes. Both models showed relative value decision biases, making them more likely to choose options with relatively better payoffs. The bias increased when the models made explicit comparisons among past outcomes, but disappeared when they estimated expected outcomes. This study has implications for understanding context-dependent choice in human agents. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at how big language models make choices. It’s like asking humans or animals which option is better – one that gave a good reward last time, or another that might give an even better reward this time. The computer models, called GPT-4 Turbo and Llama-2-70B, made lots of choices between two options, considering what happened before. They both chose the option with relatively better past outcomes, just like humans do sometimes! This study helps us understand why we make certain choices in different situations. |
Keywords
* Artificial intelligence * Gpt * Llama
