Summary of Css: Contrastive Semantic Similarity For Uncertainty Quantification Of Llms, by Shuang Ao et al.
CSS: Contrastive Semantic Similarity for Uncertainty Quantification of LLMs
by Shuang Ao, Stefan Rueger, Advaith Siddharthan
First submitted to arxiv on: 5 Jun 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 proposed Contrastive Semantic Similarity method uses CLIP to extract features for measuring uncertainty in natural language generation (NLG) tasks. Unlike previous methods that rely on logits from a natural language inference classifier, CSS extracts feature information useful for semantic clustering and uncertainty estimation. The authors apply this method to selective NLG, rejecting unreliable generations and improving the trustworthiness of large language models (LLMs). Extensive experiments with three LLMs on several benchmark question-answering datasets demonstrate that CSS outperforms comparable baselines in estimating reliable responses. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models can generate impressive text, but it’s hard to know when you should trust what they say. Right now, we don’t have a good way to measure how certain the model is about what it’s saying. Some researchers have tried using a special kind of classifier to figure out if two pieces of text are similar or not. But this approach doesn’t work well because it only looks at whether the texts match up in terms of their meaning, and it doesn’t capture all the other important details. In this paper, we propose a new way to measure uncertainty using a technique called CLIP. We show that our method is better than what’s come before for detecting when language models are unsure or making mistakes. |
Keywords
» Artificial intelligence » Clustering » Inference » Logits » Question answering
