Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
A 428-billion-parameter open-weight model achieved 59.0% on SWE-Bench Pro, outperforming GPT-5.5's 58.6%, after being compressed from 855GB to 128GB. The model's efficiency and performance mark a significant advance in A…
A new open-source Python project, code-graph-ai-summarizer, prevents LLM hallucinations in code summarization by first building a structured Code Property Graph (CPG) via static analysis tool Joern, then feeding only ver…
At the Cannes Lions International Festival of Creativity, CMOs from Bose, P&G, and other major brands are moving away from traditional agency-of-record models, instead sourcing creative work from a mix of agencies, creat…
Researchers at arXiv find that verifying coding agent outputs is now harder than generating them, as foundation models improve. They argue that no fixed reward function remains effective as policy capability grows, and v…
Researchers introduced EVOM, an agentic meta-evolution framework that uses an LLM-based design agent to automate the discovery of high-performance actor-critic architectures for reinforcement learning. EVOM outperformed …
Researchers propose a differentiable search method to discover layer-specific prompt fusion schemes for visual prompt tuning in Vision Transformers, achieving consistent gains over baselines across 34 datasets. The metho…
Researchers introduced ContextForge, a system that recycles context for long-horizon LLM inference by combining structured query generation, external memory retrieval, and controlled synthesis. In a 15-turn conversationa…
Researchers introduced AlgoEvolve, an LLM-driven evolutionary framework that generates and iteratively improves algorithmic trading strategies expressed as Python code. The system exhibits emergent regime-adaptive strate…
Researchers introduced PhyEditBench, a benchmark for evaluating physics-aware image editing in generative models, revealing significant limitations in current state-of-the-art methods. The benchmark includes 238 real-wor…
Researchers propose LCG, a framework for long-context multi-image generation that uses Sparse Relational Attention to maintain consistency across sequential outputs. The method introduces a Routing Consistency Constraint…
Researchers introduced OpenFinGym, a unified gym environment for evaluating large language model agents across multiple quantitative-finance tasks including forecasting, trading, and fraud detection. The platform address…
New research from arXiv reveals that larger language models consistently outperform smaller ones on reasoning benchmarks by an average of 6-7%, with the key advantage being constraint-guided reasoning—the ability to iden…
Researchers proposed Dynamic-dLLM, a training-free framework to accelerate Diffusion Large Language Models (dLLMs) by dynamically allocating cache budgets and calibrating decoding thresholds. The method achieves over 3x …
Researchers at arXiv found that large language models (LLMs) perform well on text-only statics problems but their accuracy drops when diagrams are added and multi-step reasoning is required. The performance decline is at…
A study comparing LLM-generated research methods against 1,000 arXiv computer-science papers found that LLMs suggest a much narrower range of methods, with provider choice showing the strongest imbalance. Academic models…
Researchers propose GeMoE, a dynamic routing method for Mixture-of-Experts-based large vision-language models that uses gating entropy to adaptively select the number of experts per token, achieving 99.5% performance ret…
A new study from Japan's AI Security Institute finds that LLM-as-judge safety evaluations are not reproducible even at temperature 0, with per-item disagreement up to 50% across runs. The researchers tested 690 API calls…
A developer building tgo, a TypeScript-to-Go compiler, reports progress on supporting Node libraries and compiling Lodash, with 122 of 316 entry points still failing. The project's complexity forced a complete overhaul o…
Researchers introduced Know2Guess, a contamination-aware multi-zone benchmark with 1,200 items across five domains to evaluate large language models' ability to distinguish answerable knowledge from abstention-expected u…
A new study on Llama 3.1 8B finds that helpfulness post-training (SFT and GRPO) significantly degrades animal compassion values compared to coding-domain post-training, with a 35.7% vs. 65.2% gap on the Animal Harm Bench…