Search and filter through extracted claims from AI researchers.
Showing 521-540 of 4537 claims
"Steering interventions targeting eval-awareness, a model's recognition that it is being tested, are increasingly used in safety evaluation pipelines, where evaluation-awareness is treated as a single quantity to be suppressed"
"Across 50 scenarios and eleven current systems, no model consistently matches the reference probabilities while recovering the range of valid behaviors."
"Recent video generation models are increasingly framed as world models."
"a world model should reproduce not only a plausible trajectory, but also the distribution of possible behaviors under the same initial observation and action"
"existing evaluations largely assess individual-video plausibility and do not test whether repeated generations recover the correct distribution"
"we test whether language prompts, initial noise sampling, or model training can reshape the model's predictive distribution"
"As large language models (LLMs) are deployed as autonomous agents, safety failures increasingly involve consequential actions."
"Using chain-of-thought (CoT) monitoring, we find that harmful execution is often preceded by intent signals in reasoning."
Post-hoc chain-of-thought labels are too coarse to show how intent changes during generation
"However, post-hoc CoT labels are too coarse to show how intent changes during generation."
"The probability of calling an intent tool provides a judge-free, fine-grained signal of the model's tendency to pursue that behavior."
Intent-as-a-tool complements CoT monitoring and expands post-hoc CoT labels into dense trajectories
"Our results show that INTENT-AS-A-TOOL complements CoT monitoring, expands post-hoc CoT labels into dense trajectories, and identifies critical steps for online intervention."
Action preferences are useful for tracking agentic misalignment during reasoning
"These findings suggest that action preferences are useful for tracking agentic misalignment during reasoning."
"this paper first identifies a performance advantage of ES over GRPO, theoretically and empirically showing that ES can lead to broader reasoning coverage, thereby better exploiting the reasoning capabilities of pretrained LLMs."
"Theoretically, we show that verifier-projected Jensen-Shannon diversity across the ES population is helpful to higher Pass@K performances."
"Empirically, unlike GRPO, which exhibits entropy collapse, ES improves Pass@1 while attaining higher Pass@K than GRPO."
"we find that despite substantial whole-model parameter drift, the task-performance gains of ES are only contributed to a sparse subset of larger-magnitude updates."
"This functional sparsity suggests that large parameter movement need not imply widespread functional change, and held-out evaluations further show that it does not necessarily lead to catastrophic forgetting."
ES requires a smaller population size in a larger LLM
"we study how hyperparameter design affects the effectiveness of ES, demonstrating that ES requires a smaller population size in a larger LLM."
"These findings position ES as a distinct reasoning post-training paradigm rather than a less effective, memory-efficient alternative to GRPO."
About $4.4K is sufficient to reach the performance of Qwen2-1.5B based on the Puro Cost Scaling Law
"the fitted law suggests that about \$4.4K, less than \$5,090, is sufficient to reach the performance of Qwen2-1.5B"
Pipeline data may be stale or degraded.
Last synthesis: 2026-09-20. 8,952 pending.