Metraipolis › archive › dc_fa4e0508d4b8

U-Space: Uncovering When and Why Uncertainty Arises in Language Models — arXiv:2610.09087

Source: arXiv (en)
Original: https://arxiv.org/abs/2610.09087
License (as archived): CC0-1.0
Archived: 2026-10-08T11:10:33.878Z
SHA-256 of the archived text: 561e0fa3e539d6d84217915b2e5cf9a08caddc3522d084ca12dbe1cf4b895808
Stable URL of this page: https://metraipolis.com/documento/dc_fa4e0508d4b8

How to cite

U-Space: Uncovering When and Why Uncertainty Arises in Language Models — arXiv:2610.09087 — arXiv (en), https://arxiv.org/abs/2610.09087 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_fa4e0508d4b8 (Metraipolis archive dc_fa4e0508d4b8, sha256 561e0fa3e539d6d84217915b2e5cf9a08caddc3522d084ca12dbe1cf4b895808)

BibTeX

@misc{dcfa4e0508d4b8,
  title = {U-Space: Uncovering When and Why Uncertainty Arises in Language Models — arXiv:2610.09087},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.09087},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_fa4e0508d4b8; sha256 561e0fa3e539d6d84217915b2e5cf9a08caddc3522d084ca12dbe1cf4b895808},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.09087. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Large language models are informing decisions with ever-higher stakes. As the consequences of their errors grow, a central question becomes harder to ignore: how much can we trust an individual answer? Yet recognizing when to defer remains difficult because language models can present incorrect conclusions with fluent explanations and an authoritative tone. Uncertainty quantification seeks to address this disconnect by estimating the reliability of individual predictions. However, many existing methods require repeated generations or separately trained components, and their scalar estimates do not reveal where uncertainty arises or how it evolves during reasoning. Recent work has also shown that generation length can be strongly associated with uncertainty estimates and correctness, raising the question of how much of an estimator's predictive power comes from uncertainty-specific information rather than output length alone. Mechanistic interpretability offers a way to address these limitations by connecting human-interpretable concepts to intermediate model states. Building on this capability, we introduce the U-Space, a low-dimensional subspace that makes a model's evolving uncertainty measurable and interpretable. We identify semantic anchors for doubt and certainty, map their unembedding directions back into the residual space, and combine their contrasts into an orthogonal basis. The U-Lens projects each token state onto these basis vectors, yielding an interpretable token-level uncertainty map that can be inspected directly or aggregated into a scalar uncertainty score. Our approach requires no correctness labels, repeated generations, or training. Across reasoning benchmarks, its confidence score outperforms established baselines under both standard and length-controlled evaluation and transfers more reliably than supervised estimators. Code: this https URL .

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI); Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2610.09087 — 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 2026-10-08T10:14:06.046Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).