Summary of Are Ppo-ed Language Models Hackable?, by Suraj Anand and David Getzen
Are PPO-ed Language Models Hackable?
by Suraj Anand, David Getzen
First submitted to arxiv on: 28 May 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Cryptography and Security (cs.CR); 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 The paper explores methods for aligning language models to remove undesirable behaviors. It focuses on positive sentiment language generation in a controlled setting, using a statically learned sentiment classifier instead of online training based on human feedback. The study examines a pretrained GPT-2 model before and after proximal policy optimization (PPO) is applied to promote positive sentiment responses. The insights gained are used to “hack” the PPO-ed model to generate negative sentiment responses and alter its weights. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper looks at how language models can be trained to say nice things. It uses a special tool that doesn’t need human feedback to make sure the words generated are positive. They take an existing AI model, GPT-2, and try to change it so it says negative things instead. This helps us understand how we can control what our language models say. |
Keywords
» Artificial intelligence » Gpt » Optimization
