Summary of Contrastive Perplexity For Controlled Generation: An Application in Detoxifying Large Language Models, by Tassilo Klein et al.
Contrastive Perplexity for Controlled Generation: An Application in Detoxifying Large Language Models
by Tassilo Klein, Moin Nabi
First submitted to arxiv on: 16 Jan 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: 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 This research paper investigates the challenge of generating undesirable and factually incorrect content by large language models (LLMs). To address this issue, the authors propose a contrastive learning objective for fine-tuning LLMs to edit implicit knowledge and generate controlled text. By optimizing this objective, the model aligns text perplexities in a contrastive fashion, allowing it to learn from itself without human labels. The authors demonstrate the effectiveness of their approach in the domain of detoxification, achieving a significant decrease in toxic content generation while maintaining general utility for downstream tasks such as commonsense reasoning and reading comprehension. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps us understand how computers can be trained to create text that is accurate and not offensive. The researchers developed a new way to train large language models (LLMs) so they don’t produce unwanted or false information. They use this technique to “edit” what the LLMs already know, making them better at generating text on their own without human help. The results show that this approach is successful in reducing the amount of harmful content generated while still allowing it to be useful for other tasks like understanding everyday situations and reading. |
Keywords
* Artificial intelligence * Fine tuning
