A Nash Equilibrium Framework for Training-Free Multimodal Step Verification
Amit Sharma, Vineeth N. Balasubramanian · 2026
Read A Nash Equilibrium Framework for Training-Free Multimodal Step Verification by Amit Sharma, Vineeth N. Balasubramanian (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-2605-20033.txt.
Description
Rohit Sinha and five co-workers verify reasoning steps of multimodal language models without training a critic. Current training-free methods just average scores from varied sources, ignoring what disagreement among those scores reveals. Their approach treats specialised judges as players in a Nash equilibrium game, where agreement signals a valid step and disagreement flags instability. A closed-form equilibrium score then filters and ranks steps by stability, beating baselines across six benchmarks by a few percent and rivalling learned critics.
Catalog record
| CREATOR | Amit Sharma, Vineeth N. Balasubramanian |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 58810 |
| LICENSE | http://creativecommons.org/licenses/by-sa/4.0/ |
| RIGHTS | CC BY SA |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.20033