Summary of Aligning Language Models with Human Preferences, by Tomasz Korbak
Aligning language models with human preferences
by Tomasz Korbak
First submitted to arxiv on: 18 Apr 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 The thesis explores approaches to aligning language models with human preferences, which can manifest as generating offensive content or perpetuating social biases. The author argues that aligning LMs can be seen as Bayesian inference: conditioning a prior (base, pretrained LM) on evidence about human preferences. The study investigates finetuning techniques using feedback given by a scoring function and distribution matching. It also examines extending distribution matching to conditional language models. Finally, the research shows that involving human feedback from the start can be more effective than using it only during supervised finetuning. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about making sure language models behave like humans want them to. Right now, they can generate mean or false things and perpetuate bad social behaviors. The author of this study wants to change that by training the models on what people think is good or bad behavior. There are a few different ways to do this, but basically it involves giving feedback to the model so it knows what people like and don’t like. |
Keywords
» Artificial intelligence » Bayesian inference » Supervised
