Tag
作者实现了一个基于 conformal risk control 的写操作安全闸门 gate-calibrated，在 tau-bench retail 上验证了风险上界成立（α=0.02 时 1.7% 错误写入未询问，达标），但闸门在每错误写入避免率上并未优于简单地每次询问用户。诊断显示模型的 verbalized confidence 和自采样一致性信号均已饱和（错误步骤上置信度也≥0.9），真正起作用的特征只是对话步数位置，说明该方案在易基准上也难以产生有区分度的置信信号。
这篇论文通过分层诊断研究探讨了交通预测精度提升为何未能改善信号控制决策，发现预测价值取决于时间可观测性、动作可识别性、动力学一致性和目标对齐，并提出了一套实用的诊断协议。
The paper studies zero-shot cross-subject continuous valence-arousal regression from synchronized EEG-fNIRS data, decomposing affect into a stimulus-shared component and an individual component calibrated via label-free alpha-band cross-channel synchrony, outperforming EEGNet and ASAC-Net baselines while reporting a systematic negative-result search over alternative architectures and features.
The paper shows that PRM-Pruned Fragment Grafting (PPFG), a cost-minimal inference-time cross-trajectory transfer technique, is statistically indistinguishable from independent parallel CoT across three reasoning LMs and six benchmarks, with the inertness traced to poor injection targeting and selection-dominated pruning dynamics.
The paper examines correctness-gated multi-teacher distillation, finding decision shifts, lost label functionality, and an inconclusive grounding audit, with no incremental benefit over hard filtering.
The paper investigates injecting frozen relational transformer embeddings into an LLM via soft tokens, reporting negative results due to performance instability and sensitivity to serialization formats.
A preregistered stress test on a small transformer shows that while latent causal structure can be localized, releasing it into behavior fails: the gate detector inverts out-of-distribution and linear release directions are bounded below sufficiency, dissociating localization from behavioral release.
A researcher made a video about their first ML research attempt on pruning vision-language models, including negative results, and submitted a paper to arXiv.
This paper presents a combined proof-of-mechanism study of ontology-amplified distillation for sovereign enterprise language models and a contextuality-audit method, using a Qwen3.6-27B student adapted via supervised fine-tuning and DPO. The results are underpowered and negative, showing no superiority over frontier baselines and zero contextuality in routing.