Reasoning as Compression: Unifying Budget Forcing via the Conditional Information Bottleneck
Fabio Valerio Massoli, Andrey Kuzmin, Arash Behboodi · 2026
Read Reasoning as Compression: Unifying Budget Forcing via the Conditional Information Bottleneck by Fabio Valerio Massoli, Andrey Kuzmin, Arash Behboodi (2026) 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-2603-08462.txt.
Description
Massoli and colleagues recast chain-of-thought reasoning as lossy compression under the conditional information bottleneck. Naive bottleneck reasoning fails for transformers, since attention breaks the Markov property across prompt, trace and response. Their fix treats the trace as a bridge holding only response information missing from the prompt. This yields a learning objective raising reward while compressing outputs under a trace prior. A semantic prior prices tokens by surprisal. Tests claim it prunes reasoning and keeps fluency and logic, improving accuracy at modest compression with losses under heavy compression.
Catalog record
| CREATOR | Fabio Valerio Massoli, Andrey Kuzmin, Arash Behboodi |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 56535 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.08462