Summary of Comparing Bottom-up and Top-down Steering Approaches on In-context Learning Tasks, by Madeline Brumley et al.
Comparing Bottom-Up and Top-Down Steering Approaches on In-Context Learning Tasks
by Madeline Brumley, Joe Kwon, David Krueger, Dmitrii Krasheninnikov, Usman Anwar
First submitted to arxiv on: 11 Nov 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 presents a case study comparing two distinct approaches to interpretability in large language models (LLMs): “bottom-up” function vectors (FV) and “top-down” in-context vectors (ICV). The authors evaluate the effectiveness of these methods for steering LLMs toward desired behaviors, highlighting their strengths and limitations. Specifically, they find that ICVs outperform FVs in behavioral shifting tasks, while FVs excel in precision-based tasks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This study compares two ways to make large language models behave in a certain way. The researchers test two methods: function vectors (FV) which are like building blocks for the model, and in-context vectors (ICV) which are more about how the model is used. They find that one method works better than the other depending on what task they’re doing. For example, one method is good at making the model change its behavior suddenly, while the other method is better at making it be very precise. |
Keywords
* Artificial intelligence * Precision
