Search and filter through extracted claims from AI researchers.
Showing 3021-3040 of 4537 claims
Context-reading models outperform term-matching lexicons for grievance detection when evaluated on non-circular benchmarks
Experience Distillation retains at least 64.8% of the gains from in-context learning across software-engineering tasks and text-adventure games without requiring further environment interaction
Latent reasoning in language models may not function as an actively consulted internal scratchpad during inference after reinforcement learning
In chess-playing models trained with RL, substituting or adding noise to latent thought vectors leaves performance unchanged, while ablating them causes only mild degradation
Reinforcement learning enabled chess models to eliminate checkmate confabulation entirely and increase legality from 48% to 61%
An animacy circuit exists in LLMs as a causal mechanism responsible for handling animacy distinctions
The animacy circuit is less localized than other known circuits and generalizes only partially across models and tasks, confirming the distributed and context-dependent nature of the animacy concept
Deep neural network training exhibits a weight-norm criticality that is distinct from learning-rate criticality, caused by the interaction between normalization and weight decay
Excessive weight decay drives scale-invariant weight norms past a critical boundary and destabilizes training, which explains why weight penalties cannot be made arbitrarily strong despite improving generalization
Existing Indonesian cultural commonsense benchmarks fail to capture cultural nuances because they evaluate LLMs on short, isolated prompts rather than dialogic contexts where culture is actually lived
CultureTalk-ID is the first dialogue-based benchmark for cultural commonsense in Indonesian and local languages, comprising 4,496 culturally grounded dialogues across 11 languages
Traditional wearable health signal analysis approaches are constrained by rigid analytical frameworks and limited personalization
LLM agents create new opportunities for Personal Health Agentic Analysis where health insights can be generated adaptively and in context
There is currently no open-source locally deployable platform capable of processing personal health data in real time while preserving privacy
In-context learning offers highly sample-efficient learning for agents but its gains disappear once experience is removed from the context
Traditional grievance detection lexicons achieve inflated performance due to circular evaluation on pools enriched with the lexicon's own selections
REFACT improves grounding by training models to decide when reasoning steps need contextual grounding and at what granularity facts should be restated
LLM reasoning traces drift away from supplied context when evidence is sparse, noisy, or conflicts with parametric knowledge
Prior ML approaches for APT detection require heavily curated training data and sophisticated preprocessing pipelines with substantial domain expertise
CAPTAIN enables robust APT detection using pre-trained language models with minimal domain-specific preprocessing, reducing engineering costs
Pipeline data may be stale or degraded.
Last synthesis: 2026-09-20. 8,991 pending.