Every processed story in chronological order, with the newest coverage first. Filter by tag, source, or score to drill in.
WebChallenger demonstrates that near-frontier web agent performance is achievable with open-weight models at a fraction of the inference cost of proprietary reasoning systems, by addressing architectural gaps rather than scaling model size.
LakeQA exposes a significant performance gap in frontier LLMs — including GPT-5.2 at 18.37% exact-match — on tasks that require jointly searching a massive heterogeneous data lake and performing multi-hop reasoning, a combination absent from prior comprehensive benchmarks.
DeLM demonstrates that decentralizing multi-agent coordination through a shared verified context can simultaneously improve benchmark performance and cut per-task cost, addressing a structural scalability bottleneck in LLM test-time reasoning.
The architecture demonstrates that constraining LLM involvement to structured front-end parsing — rather than solver code generation — can achieve high reliability on finite element simulation benchmarks while avoiding the code-correctness risks of open-ended autonomous generation.
The paper demonstrates that difficulty and consequence are approximately orthogonal signals, meaning existing difficulty-based compute routing systematically under-protects high-stakes software engineering tasks — a gap the proposed scheduler directly closes.
SMAC-Talk provides the research community with an open, structured benchmark for evaluating how LLMs coordinate, communicate, and resist deception in cooperative multi-agent environments — conditions increasingly relevant as LLMs are deployed alongside other AI agents.
Alem makes multi-agent coordination a measurable, distinct bottleneck — separate from single-agent capabilities — for the first time in a long-horizon, open-ended setting, providing a controlled testbed for developing agents that communicate, allocate roles, and execute shared plans.
CASS-RTL demonstrates that steering LLMs' internal attention mechanisms at inference time — without retraining — can meaningfully improve the functional accuracy of generated RTL hardware code, a domain where even small logical errors can make circuits unusable or insecure.
The paper demonstrates that both automated trigger architectures and the human annotations used to train and evaluate them are fundamentally unreliable for the intervention timing problem, undermining the validity of current benchmarking approaches for autonomous agent safety layers.
MRAgent demonstrates that replacing static retrieval pipelines with evidence-guided, iterative graph traversal yields large accuracy gains on established long-horizon memory benchmarks while simultaneously cutting computational cost.