AI Research news and analysis on Web Pulse: 21264 curated articles tracking the latest AI Research developments, tools, and research, updated continuously from vetted sources.
A new benchmark called ReFigBench, built on 1,000 real overview figures retrieved from arXiv papers, evaluates multimodal coding agents on reconstructing scientific figures as editable PowerPoint slides. Testing ten conf…
Applying reinforcement learning with verifiable rewards to the GLM-4-Voice speech model raised free-form accuracy on the GSM8K math benchmark to 74.8%, a new state of the art for speech-native models, according to an arX…
Researchers proposed STRETCH (Self-Taught Reasoning Evolution via Targeted CHallenge), a unified framework that uses a dynamic Stretch Zone mechanism to align question difficulty with an LLM's evolving solving capability…
Researchers introduced PersonaPath, a benchmark pairing 2,000 fine-grained learner personas with a hierarchical knowledge graph of 347 textbooks, 1,751 units, and 4,092 concepts across 77 subjects, to evaluate Knowledge-…
Researchers proposed EviGen, a three-layer framework for verifiable clinical rationale generation that uses a patient-conditioned retriever with learnable queries, an LLM generator, and a process-supervised verifier, acc…
A new paper titled "EvolveTrade: Experience-Driven Policy Refinement for Self-Evolving LLM Trading Agents" addresses the limitation that large language model trading agents rely on static hand-written tool-use policies f…
Researchers introduced VC-Attention, a method combining value smoothing and softmax casting to enable low-bit attention for Diffusion Transformers used in video generation. The work targets the dominant deployment cost o…
Metermaid AI scored 341,054 public coding-agent trajectories from 11 public datasets, including SWE-smith, SWE-rebench, Open-SWE-Traces, mini-coder-trajs and agentic-coding-trajectories, for loops, blind retries, oversiz…
A developer argues that IBM's promoted 'Chunkless RAG' approach, in which an AI agent navigates parsed document structure instead of retrieving fixed-size embedded chunks, is oversold for real-world production corpora. T…
Reinforcement learning has become the standard final stage of training frontier language models, with GRPO on verifiable rewards now the standard recipe since DeepSeek-R1 popularized it, according to a technical analysis…
ProgramDistill, a new benchmark for evaluating coding agents, tasks agents with inferring desired behavior from working software and implementing it in an incomplete application, rather than relying on issues or instruct…
David AI released DAI-S2S-ST, a single-turn speech-to-speech human preference leaderboard built on 153,000 comparative human preference ratings across 7 models, 819 distinct prompts, and 12 questions. The leaderboard ask…
Jiacheng Miao and James Zou published Paper2Agent in Nature on September 16, a six-stage multi-agent pipeline that converts a scientific paper, its code and supporting data into a Model Context Protocol server that codin…
Nunchux AI released VC-Attention, a training-free low-bit attention kernel for video Diffusion Transformers that delivers up to 3.58x attention speedup and 1.70x end-to-end speedup on the RTX 5090 at 4-bit precision. The…
Independent evaluator Vals found that Google's Gemini 3.8 Flash scored 71.7% on BioMysteryBench's human-solvable tasks and 21.6% on its hard tasks in production runs, versus the 88.8% and 56.5% Google reported in the mod…
Periodic Labs disclosed in a September 15th engineering post that its Neon 1-trillion-parameter model's final training run peaked at 1,300 Nvidia H200 GPUs, achieving over 95% cluster utilization, 4.1x the training throu…
A recursive AI system that builds its own scientific instruments and populates them with hundreds of AI agents distilled several core design principles governing hierarchical metamaterials failure, according to the resea…
AMD published a step-by-step recipe for reproducing its MLPerf Inference v6.1 submission results on AMD Instinct MI355X, MI350X, and MI350P GPUs, marking the company's fifth consecutive round of MLPerf Inference particip…
Agnes AI released Agnes-3.0-Flash Preview, a 33-billion-parameter open-weight multimodal language model under an Apache 2.0 license with a 262,144-token context window and a hybrid attention architecture. Of the model's …
Google researchers and academic collaborators described Dream RSI, a method that lets an AI system improve its own strategy for guiding scientific discovery by simulating past research history rather than running new exp…