Summary of Focus on This, Not That! Steering Llms with Adaptive Feature Specification, by Tom A. Lamb et al.
Focus On This, Not That! Steering LLMs With Adaptive Feature Specification
by Tom A. Lamb, Adam Davies, Alasdair Paren, Philip H.S. Torr, Francesco Pinto
First submitted to arxiv on: 30 Oct 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); 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 The paper introduces Focus Instruction Tuning (FIT), a novel approach to training large language models (LLMs) that enables them to condition their responses by focusing on specific features while ignoring others. This allows for adaptively steering the model’s behavior at inference-time, which can improve robustness and mitigate social bias. The authors demonstrate the effectiveness of FIT across several experimental settings, showing its ability to generalize under distribution shift and to new unseen features. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper talks about how language models are trained to do different tasks, but they often use old patterns from their training data that aren’t helpful. They came up with a new way called Focus Instruction Tuning (FIT) that lets them focus on the important parts of the data and ignore the unhelpful parts. This helps make the model’s answers better and fairer. The authors tested this method and showed it works well in different situations. |
Keywords
» Artificial intelligence » Inference » Instruction tuning
