Search and filter through extracted claims from AI researchers.
Showing 241-260 of 587 claims in topic "agents"
Budget-constrained confidence-scheduled restricted responses (CS-RNR) is the first opponent-exploitation method whose safety guarantee is a certificate the agent computes on its actually deployed strategy
Diffuse deviations in imperfect-information games pose challenges where binary release rules gather too little evidence while full best response to incomplete opponent models can be highly exploitable
Large language model-based multi-agent systems improve complex problem solving through task decomposition, agent specialization, information exchange, and intermediate validation.
Existing multi-agent systems typically treat communication topology as a fixed design choice or an offline optimization target, which is a limitation.
MANTA enables communication structures in multi-agent systems to self-evolve at inference time, adapting topology dynamically during deployment.
This is the first systematic study of large language models for parent-order execution in algorithmic trading
LLM agents acting as autonomous merchants fabricate product attributes in a majority of listings across models, even under instructions to be honest
CARP mechanism can suppress sales volume of dishonest LLM agents without requiring product-level ground truth, using only noisy complaint signals
Some LLMs (Qwen3-32B and GPT-5-Mini) can match human post-stance distributions, but only when given participants' actual initial stances
All six tested LLMs fail to simulate initial human stances themselves and to produce faithful belief updates from self-generated stances
LLMs exhibit three systematic biases when simulating human belief updates: overrepresentation of neutral positions, more frequent but smaller belief shifts than humans, and failure to rank comments accurately
Multimodal agent evaluation that reduces to final-answer accuracy cannot distinguish whether correct answers came from grounded evidence, language priors, or accidental error cancellation
Treating multimodal agent trajectories as provenance-constrained state machines with structured evidence ledgers enables better grounding verification
Computer-use agents often fail on transient GUI events because expensive autoregressive decoding is on the decision-time critical path
Adaptive Anticipatory Policy Trees improve agent success rate from 0.50 to 0.79 within contested decision windows by pre-computing conditional policy trees during idle periods
Existing post-training quantization methods are poorly suited to World Action Models because they rely on open-loop objectives, homogeneous model assumptions, and calibration distributions that do not reflect deployment
QuantWAMs framework can align quantization decisions with calibration context for World Action Models through shared-basis outlier calibration, co-training-objective saliency, and fixed-intervention rollout auditing
PACE framework extends the use of LLMs in finance from what to trade to how to execute, requiring neither explicit market assumptions nor task-specific training
PACE outperforms TWAP, Almgren-Chriss, and learning-based baselines by 0.65 bps on Shenzhen Stock Exchange Level-1 data
Frontier LLMs can achieve strong performance in social deduction scenarios requiring deception, persuasion, and reasoning under information asymmetry
Pipeline data may be stale or degraded.
Last synthesis: 2026-09-20. 8,952 pending.