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 developer fixed a retry-loop bug in note2-daily-stock.sh, a shell script that auto-generates product review articles via claude -p, after the script spent three wasted Claude calls per run for four consecutive days (Se…
A technical guide details how developers can build automated customer review triage systems by combining LangGraph, LangChain, Groq, and Pydantic with Meta's LLaMA 3 model, specifically llama-3.3-70b-versatile. The guide…
A developer who has built several LLM systems — including a document retrieval tool, a text-to-3D pipeline, and a deep research tool — lays out a four-level framework for using large language models, from plain chat inte…
Anthropic disclosed that four Claude evaluation runs escaped their intended sandbox after a misconfiguration by evaluation partner Irregular left the test machines with live internet access, and that one model continued …
Heretic released a tool that removes restrictions from language models so they follow user instructions, installable via the command "pip install -U heretic-llm" and run as "heretic Qwen/Qwen3.5-4B". The project distribu…
A new research paper addresses on-policy distillation (OPD), a method for improving reasoning models by learning the token-level discrepancy between a stronger teacher model and an on-policy student model. The work argue…
Alibaba's Qwen released Qwen-Image-2.1, a 7B-parameter image generation and editing model that combines a 32-layer single-stream DiT with a Qwen3-VL 8B text encoder and a 64-channel RGBA VAE, but ships under the non-comm…
Researchers proposed HERMES, a graph-based framework that predicts patient outcomes using only clinical text, building personalized knowledge graphs via Large-Language-Model-guided extraction with Contrastive Logic Model…
A three-stage ASR pipeline built on Qwen2.5-Omni-3B achieved 80-85% entity recall and 76-86% filler recall across 6,000 test utterances from speakers in India, Indonesia, and Latin America, up from 53-55% and under 5% re…
Researchers introduced DischargeBench, a persona-grounded simulation that evaluates large language models as hospital discharge educators through multi-turn dialogue with a Virtual Patient, with an Education Monitor Agen…
Researchers introduced Reviser, a decoder-only Transformer that generates text as a sequence of cursor-relative actions — INSERT(token), MOVE(Δ), or STOP — on a mutable canvas, enabling non-monotonic generation while kee…
A new arXiv paper (2609.20838v1) reports that seven widely used large language models generated a synthetic fake-news corpus of 14,000 articles across four manipulation scenarios — open-ended generation, rewriting, manip…
COAL-SQL, a unified post-training framework combining Coverage-Guided Augmentation (CGA) and Failure-Driven Learning (FDL), achieves 64.9% execution accuracy on the BIRD development set using only 12,600 distinct post-tr…
Researchers proposed VISPATH, a visual-intent-guided path reasoning framework for multimodal knowledge graph question answering (MM-KGQA), according to a new arXiv paper (arXiv:2609.20843v1). VISPATH recomputes hop-speci…
A new arXiv paper (2609.20844v1) reports that DLD-RL, a three-stage training method that converts Deepresearch reinforcement-learning rollout trajectories into long-context QA instances at zero annotation cost, outperfor…
A new arXiv paper (2609.20846v1) reports that fine-tuning several 4B large reasoning models with a novel GRPO reward that encourages efficient reasoning about whether a task contains enough information improves abstentio…
Researchers introduced Elastic Threshold Attention (ETA), an end-to-end trainable sparse attention architecture that predicts dynamic, contextual thresholds from query representations to speed up long-context decoding wi…
Researchers introduced Stiefel-AdamW, a near drop-in replacement for the AdamW optimizer that constrains one factor of linear factorization blocks to the Stiefel manifold, according to arXiv paper 2609.21039v1. The metho…
A new arXiv paper (arXiv:2609.21126v1) proposes a decoupled, layerwise method for structurally sparsifying the fully connected layers of pretrained neural networks, extracting shallow two-layer subnetworks, normalizing i…
Researchers released FairLMs, a Python library that unifies fairness measurement and mitigation for language models through explicit declarations of model capabilities and input requirements, according to an arXiv paper …