Metraipolis › archive › dc_7c32dec35fdf

Metacognitive Steering: Learning the Structure of Scientific Judgment — arXiv:2609.16245

Source: arXiv (en)
Original: https://arxiv.org/abs/2609.16245
License (as archived): CC0-1.0
Archived: 2026-10-15T18:32:01.146Z
SHA-256 of the archived text: 7000c77574003922c7af5dc32d609c649ddcff8de4fde838c3c8dba41a7abd6a
Stable URL of this page: https://metraipolis.com/documento/dc_7c32dec35fdf

How to cite

Metacognitive Steering: Learning the Structure of Scientific Judgment — arXiv:2609.16245 — arXiv (en), https://arxiv.org/abs/2609.16245 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_7c32dec35fdf (Metraipolis archive dc_7c32dec35fdf, sha256 7000c77574003922c7af5dc32d609c649ddcff8de4fde838c3c8dba41a7abd6a)

BibTeX

@misc{dc7c32dec35fdf,
  title = {Metacognitive Steering: Learning the Structure of Scientific Judgment — arXiv:2609.16245},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.16245},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_7c32dec35fdf; sha256 7000c77574003922c7af5dc32d609c649ddcff8de4fde838c3c8dba41a7abd6a},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.16245. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Long-horizon scientific discovery requires agents to alternate between exploration, disciplined execution, and critical reassessment as evidence changes. Current language models are trained primarily on the products of science and optimized using outcome-level signals, providing limited supervision for these process-level shifts in scientific judgment. We investigate whether such judgment can be recovered from scientist interaction traces and used to control the internal computation of a frozen frontier model. Using contrastive interventions collected during real scientific research, we identify a coordinated, low-dimensional control structure within Kimi 2.6, a trillion-parameter mixture-of-experts model. Residual analysis, attention-weight subspace alignment, and cross-layer singular value decomposition converge on a mid-depth control surface spanning key layers. We introduce Metacognitive Steering, an inference-time controller that reads the model's cognitive regime and dynamically composes layer-specific interventions for exploration, procedural convergence, or critical reassessment without modifying model parameters. Behavioral analyses show that this control produces more sustained exploration, explicit pruning, and evidence-responsive synthesis. We operationalize the method in Columbus-1, an autonomous research system that identified eight independently reproduced, attacker-reachable vulnerabilities in BlueZ and directed the design, simulation, and fabrication of a ten-foot rocket intended to land propulsively using non-throttleable solid motors. Together, these results show that process-level scientific judgment can provide supervision for interpretable, dynamic control over a model's reasoning strategy.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-16 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.16245 — 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 2026-10-15T18:03:09.420Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-16?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).