Londoners can now hire self-driving taxis for the first time after Wayve's autonomous vehicles were added to the Uber app on Thursday, with only 15 cars licensed so far. The vehicles will have a human…
A new survey from arXiv (2505.21907v3) defines AI copilots and introduces a taxonomy of preference optimization techniques across pre-, mid-, and post-interaction phases, evaluating each technique's a…
Researchers released SpecMine, a large-scale corpus of spec-driven development artifacts from public GitHub repositories, capturing 470,795 spec.md/specs.md files across 73,030 repositories attributed…
A new multilingual summary meta-evaluation dataset (BASSE) containing human judgments on 2,040 abstractive summaries reveals that proprietary judge LLMs correlate most strongly with human ratings, fol…
Researchers introduced SABER-Math, the first fully automated benchmark for evaluating mathematical information retrieval without expert annotation, built from 283K high-school-level math problems. The…
Researchers introduced ToolGate, an executable acceptance pipeline for building scientific benchmarks that require specialist software, and demonstrated it on FEniCSx with 500 generation attempts. The…
Researchers introduced SkillGLoW (Global-Local Weave), a method for consolidating procedural skills in LLM agents, achieving a 17.2-point average gain on hard benchmarks over no-skill baselines across…
A new arXiv preprint (2609.02092v1) introduces SCOPED-Hiring, a process-aware fairness diagnosis pipeline for LLM-based hiring multi-agent systems, which logged over 311,000 structured decision trajec…
Researchers propose CHIME (Credit-Aware Hierarchical Memory Evolution), a self-evolving memory framework that separates planning and execution memory banks to address credit assignment in long-horizon…
Researchers introduced EmoStance, a method that uses emoji distributions as weak supervision to control the affective orientation of empathetic response generation, achieving a 62.2% decisive win rate…
Researchers introduced MASkills, a continual learning framework that optimizes multi-agent LLM systems through agent skills, enabling skill libraries to evolve via refinement, induction, consolidation…
A new arXiv paper (arXiv:2609.02246v1) argues that LLM-as-a-judge should be demoted from oracle to advisor in self-improving agent pipelines, citing eleven failure modes observed over months of produc…
A new arXiv paper (2609.02168v1) introduces FUSE, a modular framework for evaluating dangerous capabilities of large language models across three orthogonal pipelines—Knowledge (K), Defense (D), and H…
Researchers propose Potential-Guided Policy Optimization (PGPO), a new reinforcement learning method for multi-turn agentic tasks that estimates empirical state potentials from anchor-state-group retu…
Researchers introduced ProSE-Plan, a depth-2 Bayes-adaptive planner that improves AI assistance by accounting for users' bounded rationality and evaluability constraints, outperforming evaluability-un…
Researchers propose Prior-Guided Tuning (PGT) with Contrastive Prior Steering (CPS), a training method that uses task-level natural-language priors as auxiliary learning signals for low-resource LLM t…
Researchers propose APEx, a hierarchical experience utilization framework that organizes interaction history into instance-level trajectory memories and category-level procedural skills, coupled throu…
A new arXiv paper (2609.02264v1) introduces Codebook Agent, a method that compresses successful communication topologies for LLM multi-agent systems into a 16-entry codebook and uses a reward-weighted…
Researchers introduced CivBench, an open-source benchmark for evaluating language model agents in long-horizon, tool-mediated environments via the Model Context Protocol (MCP), featuring 300+ turn epi…
Researchers introduced SCX Router, a lightweight GLiClass-based model for zero-shot selection among LLM inference endpoints, achieving an aggregate top-1 score of 0.707 on a 1,000-task subset versus 0…