A new arXiv study introduces a Holistic Sustainability Score (HSS) to compare small language models (SLMs) and quantized large language models (LLMs) for edge AI, finding that Qwen3-30B-A3B/GGUF Q4 ra…
Researchers introduced Value Over Language Model (VOLM), a training-free framework that measures a document's original contribution by comparing LLM reconstructions from partial content against those …
Researchers from OpenBMB released ChatDev 2.0: DevAll, a no-code platform for building and executing multi-agent systems that combines a declarative executable graph abstraction with a cycle-aware exe…
A new study from arXiv (2609.00718v1) finds that structured pruning is the first stage where driving capability is lost in compressed autonomous driving policies, while distillation improves the prune…
Researchers introduced SOVER, an LLM-assisted SMT framework that formally certifies optimization reformulations using Z3 and dReal, achieving 99.33% accuracy (149/150) on the new NLEquiv-150 benchmark…
A new arXiv paper (2609.00731v1) introduces Agentic Empirical Asset Pricing (AEAP), a paradigm in which LLM agents autonomously conduct the scientific discovery process for asset pricing, and proposes…
A new study on arXiv (2609.00904v1) finds that large language models (LLMs) do not demonstrate reliable control over privileged internal representations when the control target is not inferable from t…
A new arXiv preprint (2609.01466v1) presents a live trace model that folds an agent's append-only event ledger into typed run state, enabling an LLM reader to answer monitoring questions with 14x-15x …
A new arXiv preprint (2609.01519v1) reports that an initial evaluation of marketplace guardrails in an LLM agent commerce testbed overstated welfare gains, with effects shrinking from +87.4, +35.0, an…
Researchers propose SAGE (Selective Agent Guidance via Entropy), a framework that queries a Vision-Language Model (VLM) only when a learner is uncertain and distills the guidance into a lightweight Re…
A new arXiv study (2609.00879v1) proposes a two-stage training framework combining Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO) to improve multimodal disaster damage assessmen…
Researchers propose CARE (Contrastive Anchor-based Rubric Evolution), a method that uses anchor responses from frontier models to dynamically evolve rubrics during LLM post-training, addressing reward…
Researchers introduced InteractBench, a benchmark of 322 interactive competitive programming problems from Codeforces, AtCoder, IOI, and ICPC, revealing that even the most advanced large language mode…
Researchers introduced WorldBench, a multilingual benchmark of 1,600 persona-grounded everyday tasks across seven languages and eight cultures, finding that frontier models achieve only 49.2% Constrai…
Researchers propose ARISE-RL, a self-evolution framework that couples a task/rubric Generator and a reasoning Solver through rubric-mediated co-evolution to train open-ended agents via reinforcement l…
Researchers from an unspecified institution have proposed a framework for solar-powered generative AI on satellites, addressing energy constraints by jointly optimizing computation and communication. …
Researchers introduced Latent Recurrent Thoughts (LRT), a method that enables frozen large language models to reason in continuous representation space using a small recurrent reasoner, outperforming …
Researchers released analog-db, an open-source, versioned database for analog integrated circuit designs that uses a domain-specific language to capture process-neutral topologies, reusable testbenche…
Researchers introduced a framework for evaluating behavioral fidelity in long-horizon activity simulations, finding that statistical priors bring activity and sequence distributions closest to real be…
Researchers propose H2Table, a hierarchical hypergraph-enhanced framework for complex table reasoning, which represents tables as nested hypergraphs and uses a tailored encoder with learnable query ve…