Efficient Latent Semantic Clustering for Scaling Test-Time Computation of LLMs
Sungjae Lee, Hoyoung Kim, Jeongyeon Hwang, Eunhyeok Park, Jungseul Ok · 2025
Read Efficient Latent Semantic Clustering for Scaling Test-Time Computation of LLMs by Sungjae Lee, Hoyoung Kim, Jeongyeon Hwang, Eunhyeok Park, Jungseul Ok (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-00344.txt.
Description
Sungjae Lee and four co-authors present in 2025 a faster way to group answers from large language models. When a model writes many answers to one prompt, some say the same thing in new words. Grouping those answers helps judge how sure the model is.
Older methods call on a second model to do the grouping, which costs time and can miss context. The new method, called Latent Semantic Clustering, uses hidden states inside the first model itself. Tests on many models and data sets show it cuts cost a great deal. It also matches or beats older methods, which helps when models think through long chains of steps.
Catalog record
| CREATOR | Sungjae Lee, Hoyoung Kim, Jeongyeon Hwang, Eunhyeok Park, Jungseul Ok |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 52 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2506-00344