Metraipolis › archive › dc_4aaa0817488e

Euston: Training Away Mathematical Sycophancy Without Losing the Mathematics — arXiv:2609.23205

Source: arXiv (en)
Original: https://arxiv.org/abs/2609.23205
License (as archived): CC0-1.0
Archived: 2026-10-09T06:49:46.475Z
SHA-256 of the archived text: 13b1702ac5f66b3a36defbac2863415fe1e90888518c9d6fd3ef313145d61045
Stable URL of this page: https://metraipolis.com/documento/dc_4aaa0817488e

How to cite

Euston: Training Away Mathematical Sycophancy Without Losing the Mathematics — arXiv:2609.23205 — arXiv (en), https://arxiv.org/abs/2609.23205 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_4aaa0817488e (Metraipolis archive dc_4aaa0817488e, sha256 13b1702ac5f66b3a36defbac2863415fe1e90888518c9d6fd3ef313145d61045)

BibTeX

@misc{dc4aaa0817488e,
  title = {Euston: Training Away Mathematical Sycophancy Without Losing the Mathematics — arXiv:2609.23205},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.23205},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_4aaa0817488e; sha256 13b1702ac5f66b3a36defbac2863415fe1e90888518c9d6fd3ef313145d61045},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.23205. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Reasoning language models are trained to produce solutions, not to refuse them, and this bias persists when the problem they are handed is false. Asked to prove a corrupted theorem, a strong model will typically comply and produce a confident derivation of something untrue. We present Euston, an 8B mathematical claim-verification model trained to resist exactly this. Training data were generated with GraphSynth, a probabilistic factor-graph generator that couples attribute-level diversity to decode-time structural masking and span-synchronized verification, yielding 3{,}026 matched true/corrupted statement pairs (6,052 statements) drawn from arXiv papers spanning 2010--2025. We fine-tuned DeepSeek-R1-8B with GRPO under a rule-based, zero-API reward for 189 steps on four H100 GPUs. On a balanced 200-true/200-false held-out split, balanced accuracy rises from 29.50% to 63.75% and the discrimination gap---the difference between the rate of calling false statements false and the rate of calling true statements false moves from -0.5% (z=-0.1) to +27.5% (z=+6.0). Critically, the gain is not purchased with general mathematical ability: AIME 2026 accuracy under official semantics is 65.00% against a 69.17% base, a difference of -4.17% that is not statistically significant, whereas an earlier run of the same recipe on a smaller GraphSynth corpus collapsed to 40.00%. Median response length also falls from 19,217 to 18,296 tokens and the truncation rate from 25.8% to 8.3%, so the improvement does not come from thinking longer. We report the result together with the confounds that bound its interpretation, principally the all-false composition of the official evaluation sets and the low precision implied at realistic error prevalence.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-22 in cs.CL).
Categorie arXiv: Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2609.23205 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-09T06:48:44.981Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-22?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).