Metraipolis › archive › dc_583ffb472509

What Matters for Latent Reasoning with Flow Matching — arXiv:2610.06666

Source
arXiv
Original
https://arxiv.org/abs/2610.06666
License (as archived)
CC0-1.0
Archived
2026-10-06T05:28:57.107Z
SHA-256 of the archived text
5d1ad9bd225d32dc363959f396de05ccf0a8a0828a6bd76061594f6d9da85f73
Stable URL of this page
https://metraipolis.com/documento/dc_583ffb472509

How to cite

What Matters for Latent Reasoning with Flow Matching — arXiv:2610.06666 — arXiv, https://arxiv.org/abs/2610.06666 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_583ffb472509 (Metraipolis archive dc_583ffb472509, sha256 5d1ad9bd225d32dc363959f396de05ccf0a8a0828a6bd76061594f6d9da85f73)
BibTeX
@misc{dc583ffb472509,
  title = {What Matters for Latent Reasoning with Flow Matching — arXiv:2610.06666},
  howpublished = {arXiv},
  url = {https://arxiv.org/abs/2610.06666},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_583ffb472509; sha256 5d1ad9bd225d32dc363959f396de05ccf0a8a0828a6bd76061594f6d9da85f73},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv, https://arxiv.org/abs/2610.06666. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Latent reasoning lets a large language model (LLM) think in a continuous space and verbalize only the answer. We argue that an effective latent thought must meet five requirements: it should be useful, helping produce the correct answer rather than merely changing it, diverse, so that resampling yields different reasoning trajectories, explainable, so that a decoded chain of thought (CoT) reflects reasoning the answer actually follows, refinable with more inference compute, and efficient, costing less than an explicit CoT at comparable accuracy. Current methods rarely meet these requirements: they learn shortcuts from the question, distill the explicit CoT into their weights, or imitate it one token at a time. We focus on flow matching in a learned latent space, the family we argue is best placed to meet them, and identify the training choices that make it work. The result is Flow-based Latent Reasoning (FLaRe), a simple recipe covering what the latent space encodes and how to shape it, where to train the flow, how to read out the answer, and a final stage of training on the model's own verified thoughts. A probe for each requirement shows that FLaRe improves on prior latent methods in all five. It also compares favorably with them on arithmetic benchmarks, while reaching 97% of the accuracy of explicit CoT at a quarter of its latency.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG) ; Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2610.06666 — 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 2026-10-06T05:24:20.468Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).