Test-time reasoning systems often respond to failure by restarting or revising the latest step, even when an earlier decision caused the error. We introduce CORE, a search controller that requests a certified conflict core from a verifier, backjumps to the latest decision in that core, and caches the conflict to avoid...
Si-Yu Song, Rui-Peng Xu, Jia Lin et al.· arXiv.org· 0 citations
SciCore, a dual-branch reviewer that averages a full-manuscript judgment with a judgment based on an extracted, structured science core, achieves a leading joint stability-discrimination profile among the benchmarked reviewers while maintaining competitive human alignment.
Chen-Guang Wang, Ming Li, Cheng-Rui Fan et al.· 0 citations
The authors' manual, post-hoc analysis is descriptive: in one inspected case, optical character recognition (OCR) and block grouping merged the relevant passage into another block, and the system answered from unrelated text.
Settle extends the accuracy-token-count Pareto frontier of the evaluated stopping methods and predicts whether a correct answer will remain correct in completed traces.
Ryan Brown, Zi-Hao Fu, Chris Russell· 0 citations
Reach audiences
Advertise in front of researchers, engineers, and readers.
It is proved that the clipped objective can fail to correct the student toward the teacher and can instead push clipped and unclipped token probabilities away from its teacher.
At 460 h of clean LibriSpeech, the seed moves fairness metrics more than compression does on most demographic axes, and a balanced 3x3 decomposition attributes 85.3% of the variation in Fair-Speech ethnicity normalized gap to the seed against 8.3% to compression.
Srishti Ginjala, E. Fosler-Lussier, Srinivasan Parthasarathy· 0 citations
GraphForge, an evidence-graph based framework that grounds both the task and its verification in real files, and rejects fine-tuning on the SFT model's own rollouts yields further improvements on all three benchmarks, suggesting that the rubrics provide a useful selection signal.
Qi-Sheng Su, Han-Chen Wang, Guang Zhu et al.· 0 citations
Finding a relevant paper is not the same as producing a verifiable answer from it. LitTraceQA requires canonical paper identifiers, exact evidence at the page or object level, and typed answers that match the evaluator. We call the separation between source access and scorer-visible correctness the grounding contract g...
Model-attribution classifiers can often identify which language model produced a text, making model-specific writing patterns a signal of provenance. Accurate attribution on unmodified text, however, does not show whether the prediction still identifies the original source after deliberate rewriting. We formulate this...
Hao-Han Yuan, Si-Min Chen, Xi Niu et al.· 0 citations
It is established that principled and informed team selection is critical and strong model selection algorithms for assembling effective multi-LLM systems are presented.
Zong-Wan Cao, Zi-Yuan Yang, Shang-Bin Feng et al.· 0 citations
Overlap-Unique-Conflict (OUC) extraction, a cross-narrative task that extracts all overlapping, conflicting, and unique clauses from two narratives, is introduced, finding that unique information is far easier to extract than overlap and conflict.
Steering large language models typically relies on linear, context-independent interventions in activation space, an assumption that recent work has challenged and that can induce an information bottleneck when a fixed representation must encode many behavioral distinctions. We introduce MetaSteer, a method that learns...
Flint-interactive gives AI agents a language for creating interactive visualizations. Click a chart, filter data, highlight related content, and explore beyond the view the author originally chose. The post Flint-interactive: A language for creating interactive data visualizations appeared first on Microsoft Research.
