Summary of An Assessment Of Model-on-model Deception, by Julius Heitkoetter et al.
An Assessment of Model-On-Model Deception
by Julius Heitkoetter, Michael Gerovitch, Laker Newhouse
First submitted to arxiv on: 10 May 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
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 introduction of a method to investigate complex, model-on-model deceptive scenarios is proposed in this paper. The authors create a dataset of over 10,000 misleading explanations by asking large language models like Llama-2 7B, 13B, 70B, and GPT-3.5 to justify incorrect answers for questions in the MMLU dataset. They find that when these models read the deceptive explanations, they are all significantly deceived. The study also reveals that more capable models are only slightly better at resisting deception than less capable ones. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper explores ways to detect and defend against deception in highly capable language models. It shows that even top-performing models can be tricked into producing misleading outputs, which undermines their reliability. The authors create a large dataset of deceptive explanations and demonstrate that these models are vulnerable to being misled by each other. |
Keywords
* Artificial intelligence * Gpt * Llama
