Use this view to inspect the underlying evidence corpus. For ranked developments, decision posture and interpretation, use Signals.
Raw feed or Signals?
Raw feed is chronological evidence. Signals ranks and interprets material change.
An arXiv research paper introduces benchmarking for deep search LLM agents, focusing on error analysis across multi-step execution traces.
Researchers propose Effect-Marginal Policy Optimization to improve multi-turn reinforcement learning for long-horizon LLM agents.
Researchers introduce FairAMO, a controlled evaluation protocol to disentangle the performance gains of agentic time series model orchestration.
Researchers proposed BRIDGE, a model-agnostic acceptance gate to prevent false improvements during iterative model updates.
Researchers formalise a spectral threshold governing stability in multi-agent performative prediction and feedback loops.
Researchers introduced Temporal-Context Edge Enrichment (TCEE), using time-series foundation models to improve graph-based financial crime detection.
Researchers propose post-training methods to preserve chain-of-thought monitorability in capable AI models against degradation.
An arXiv preprint argues that general-purpose LLMs overlook contextual privacy, critiqueing one-size-fits-all enterprise privacy models.
Researchers introduced FrontierCS 2, a benchmark evaluating LLM agents on 48 open-ended computer science research tasks.
Researchers propose treating LLM classifiers as noisy measurement instruments to optimize classification accuracy and query costs.
An arXiv research paper audits two AI-assisted repositories to examine how well code provenance and agent residue are recorded.
Researchers analyze how state-space foundation models simulate limit order books and interpret market microstructure dynamics.
Researchers introduced JaxMARL-BC, a framework that accelerates multi-agent reinforcement learning for macroeconomic modelling.
Researchers introduced AgentPolicyForge, a framework for creating policy benchmarks and training adaptive guards for agentic AI tools.
Researchers propose a method for scoring code edits in self-improving coding agents before execution using steering-split metrics.
Researchers proposed hyperspherical simplex representations to enable backward-compatible vector retrieval without re-indexing galleries.
Researchers analyzed 802,863 Polymarket contracts to predict trading volume and dispute risk using only listing metadata.
Researchers propose Tree-Induced Diagnostic Evaluation (TIDE) to provide fine-grained diagnostic evaluation of LLM behaviors beyond aggregate scores.
Researchers propose Code Comprehensor, combining hierarchical documentation and cross-modal retrieval for large code repositories.
An arXiv paper benchmarks post-training quantization methods for LLMs to clarify memory and computational efficiency trade-offs.
Researchers propose OracleAD, an unsupervised and interpretable framework for detecting anomalies in multivariate time series data.
Researchers introduced an interval-based AUC framework to evaluate uncertainty-aware, interval-valued risk prediction models.
Researchers propose evidential uncertainty quantification to detect hallucinations and misbehaviors in large vision-language models.
Researchers propose RelAgent, an LLM-based autonomous data scientist framework designed for relational learning tasks.
Researchers propose InfoQuant to shape activation distributions for low-bit LLM quantization, addressing deployment bottlenecks.
Researchers propose a queue control model for optimizing task routing between automated chatbots and human agents in service systems.
A recent research paper examines feature superposition in neural networks, explaining polysemantic neurons and scaling laws.
Researchers identify principle-level search collapse in feedback-driven LLM optimizers, where search becomes confined to narrow design principles.
Researchers propose optimized conformal selection methods to improve selective inference and control false discovery rates in data analysis.
A research study on arXiv evaluates adversarial random forests for generating synthetic data to address privacy and access limits.