#01Oct 22, 2026
cs.LG
Questioning Instrumental Convergence in RLVR: Checkpoint Evidence from OLMo-3
Barış Alparslan, Sandy Tanwisuth, Caleb Petterson
Instrumental convergence is often invoked as a general reason to expect capable agents to acquire or express power-seeking subgoals: self-preservation, resource acquisition, deception, or resistance to oversight which poses an important threat to the society. For LLMs, reinforcement learning is the first stage of post-training that directly optimizes behavior against an objective; therefore, it is the stage where convergent instrumental behavior, on IC grounds, likely to appear. Does reinforcement learning with verifiable rewards (RLVR) make a model more likely to express instrumental-convergence behavior, or does this depend on the training regime, evaluation frame, and stage of post-training? We study this question as a checkpoint-conditioned behavioral trajectory rather than as a single before/after safety score. Using InstrumentalEval, we evaluate OLMo-3-7B-Think across SFT, DPO, 13 intermediate RLVR checkpoints, and the final RLVR model. The trajectory is nonmonotonic: instrumental rate rises from 25.0% at SFT to 46.1% near RLVR step 200, then falls to 11.8% at the final checkpoint. Paired item-level tests and step-wise regression support the final SFT-to-RLVR reduction. Prompt framing does not explain this pattern: a harmful-prompt indicator combined with other framing effects explains only 15% of within-checkpoint score variance on average, and Iterative Null-space Projection (INLP) removes a detectable framing direction without changing the SFT-to-final-RLVR gap. Together with preliminary 32B endpoint results (35.53% SFT to 22.37% RLVR final), these findings indicate that RLVR does not monotonically increase instrumental behavior in this setting: the rate rises early, then falls below the SFT baseline. The result is specific to this model family and benchmark, and judge agreement is moderate, so absolute rates should be read cautiously.
#02Oct 22, 2026
cs.AI
Who Is Sycophantic to Whom? Confident LLMs, Human Deference, and the Collapse of Trust Calibration
Deepthika S, Pawni Yadav, B. Sankar
Research on sycophancy asks whether large language models (LLMs) bend towards what their users want to hear. We ask the reverse question: when an LLM is confident, consistent, and wrong, do users bend towards it? In a within-subjects experiment, 32 participants learned up to five science topics by chatting with five assistants built on the same model (GPT-5-mini). Unknown to participants, each assistant's briefing had 0, 25, 50, 75, or 100% of its facts replaced by plausible falsehoods, and every assistant spoke with full confidence. A surprise quiz after each topic offered the correct fact, the planted falsehood, and two distractors, with a confidence rating for every answer. Across 104 quizzes and 1,040 answers, scores fell from 7.4 to 2.9 out of 10, and the planted option was chosen on 60% of falsified items against 11% of correctly taught ones (odds ratio 12.5). Transcript coding showed that whenever the assistant stated a falsehood, it was adopted about 65% of the time at every falsification level. Absorbed falsehoods were held with high confidence (8.4/10), the modeled confidence advantage of right over wrong answers vanished, and the discrimination of confidence for correctness fell from an AUROC of 0.70 to chance, while post-task trust ratings barely moved. Participants tended to return to the chat more when the assistant was worse, but returning went with more absorption, not less, and 30% of all turns pasted quiz questions back to the source being checked. The assistant itself was almost never sycophantic and never conceded under pushback. We introduce a baseline-corrected Human Deference Index to quantify this human-side sycophancy and argue that trustworthiness is a property of the human-AI pair, that sycophancy evaluations should measure both directions, and that learning interfaces must make reliability legible claim by claim.
#03Oct 22, 2026
cs.AI
What Makes a Good Unmasking Policy? A Controlled Study of Learned Unmasking in Diffusion Language Models
Jinya Sakurai, Xun Xu, Jaehong Yoon
Masked diffusion language models generate tokens in parallel, but their efficiency depends on which predictions are committed. Learned unmasking policies improve the speed--accuracy trade-off, yet differ in supervision, inputs, and capacity, obscuring which choices matter. We view supervised policies as distilling a decoding procedure through a teacher and examine independently what the teacher labels as committable, on which states the policy imitates it, what the policy observes, and how much the number of function evaluations (NFE) must decrease to offset policy overhead. We find that labeling predictions matching the final output is overly permissive at low NFE, and hidden states are weak alone but can complement confidence. About a 10% NFE reduction suffices to offset policy overhead for policies up to 7.5M parameters across tested settings. We therefore train a policy on trajectory-contraction targets, which accept only commitments staying on a reference trajectory, using confidence and hidden states. On LLaDA-8B-Instruct, the resulting 856K-parameter policy achieves the best accuracy--NFE frontier and highest throughput on GSM8K, is competitive at moderate targets on MATH-500 and HumanEval, and largely transfers to Dream-7B-Instruct.
#04Oct 22, 2026
cs.CV
MoG-VLM: Efficient Video Anomaly Detection through Motion-Guided Vision-Language Models
Khaled Waleed Dawoud, Muhammad Zaigham Zaheer, Khan Muhammad and 1 more
Vision-Language Models (VLMs) have shown strong potential for video anomaly detection, but dense frame-level inference makes them impractical for real-time deployment. Existing acceleration commonly relies on uniform sampling, which ignores content dynamics and can either waste computation on static segments or miss localized anomalies. We propose MoG-VLM a motion-adaptive anomaly detection framework that leverages motion purely as a lightweight control signal to decide when VLM inference is necessary. To this end, we introduce Temporal Gradient Magnitude (TGM) which performs content-aware sampling to concentrate computation on informative regions while avoiding redundant processing. We further introduce a hierarchical dual-stream design in which dense multi-frame VLM reasoning is activated only around frames with elevated anomaly likelihood, via anchor-triggered context-window expansion with early termination. A fusion module then reconciles these instantaneous semantic cues with temporal action dynamics. Across four benchmarks, our method achieves 91.91%, 84.30%, 87.88%, and 78.96% AUC-ROC on XD-Violence, UCF-Crime, MSAD, and ShanghaiTech, respectively, establishing new state of the art on three benchmarks under the zero-shot, training-free setting. On XD-Violence, we improve over the previous best result by +0.57% while delivering a 1.31x inference speedup. These results demonstrate that content-aware allocation of VLM capacity, processing fewer frames more intelligently rather than all frames uniformly, is a more effective axis of progress than uniformly densifying VLM inference across frames. Code is available on GitHub: https://github.com/kdawoud91/MoG-VLM.
#05Oct 22, 2026
cs.AI
Jev-DLC: Decision-Level Continual Learning from Typed Feedback
Yangyang Feng, Jianle Gan, Zhuoyan Feng and 1 more
Individual preferences often become visible only after deployment, when a user accepts one decision or revokes it in favor of another. Jev-style typed interfaces make this feedback explicit by identifying the selected action and, under revocation, the decision it replaces. We study how these consequences persist across requests as Decision-Level Continual Learning from Typed Feedback (Jev-DLC). Jev-DLC treats each correction as both evidence about a recurring action identity and a directed relation between the preferred and rejected choices. It preserves these signals in complementary bounded channels above a frozen scorer and combines their scores to shape the next decision. The method retains no raw interaction history, adds no scorer call, and performs no runtime language-model backward pass. We also introduce TCPA-Bench, a chronological suite that credits a correction only through its effect on later decisions. On TCPA-Repeat, Jev-DLC reaches 44.83% user-macro accuracy overall, 2.43 points above Recent-4 history replay. After at least two earlier feedback events, it reaches 56.07%, a 7.54-point margin over the strongest reported comparison at this depth. These results show how explicit typed corrections can become compact causal state for continual personalization.