Every processed story in chronological order, with the newest coverage first. Filter by tag, source, or score to drill in.
The survey provides the first structured taxonomy of Multimodal Code Intelligence, connecting mature code-generation benchmarks to emerging agentic settings and identifying verification gaps that current text-to-code evaluations do not address.
The findings show that agent+tool evaluations cannot assume the agent adds judgment on top of the tool — and that the gap between parrot behavior and optimal action widens, not shrinks, as LLM capability scales.
ASSAY demonstrates that matching skills to tasks at inference time — rather than global library curation — is the key bottleneck for experience-based agent improvement, achieving state-of-the-art results on two benchmarks without any weight updates.
DeepRoot is the first system to simultaneously achieve low hallucination rates (7–10%) and high reasoning coherence on historical medical text, demonstrating a viable path for converting pre-ontological archives into verifiable drug-discovery leads at scale.
RetailBench exposes that current LLMs cannot sustain coherent long-horizon decision-making in economically grounded environments, with most models failing to complete even a 180-day simulation and all falling substantially short of an oracle policy on net worth and sales.
LatentGym fills a gap left by existing frameworks by providing the first controllable latent structure and disentangled exploration/exploitation metrics for measuring cross-task experiential learning in LLM agents.
CSTS addresses a core bottleneck in agentic LLM development by replacing manual skill engineering with an automated, multi-model collective process that explicitly tests whether skills transfer across models — a property the paper identifies as critical for robust generalization.
ACCORD demonstrates that a training-free grounding layer can close a substantial portion of the task-completion gap in LLM agents across both digital and embodied benchmarks, without modifying the underlying model.
The contrastive context-selection objective demonstrably outperforms simply adding more contrastive data, showing that how the training signal is structured — not just what data is used — drives grounding improvements in both agentic and multimodal LLM settings.
Despite code access giving LLM agents a measurable edge on time series tasks, a 22–34% error rate on benchmark questions exposes a concrete reliability gap that limits their use in high-stakes automated decision-making domains like finance and healthcare.