Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Databricks and the Linux Foundation announced on June 10, 2026 that Delta Sharing has evolved into OpenSharing, an open, vendor-neutral protocol that extends zero-copy sharing beyond tables to agent skills, AI models, an…
Deepsense.ai released an October update to its EDA Benchmark, which runs language models and agentic systems five times each across 10 synthetic exploratory data analysis tasks to measure both average quality and run-to-…
A developer built Log Doctor, a Kubernetes CronJob that runs every 15 minutes on EKS, pulls aggregated error logs from OpenSearch via Fluentd, sends a bounded, redacted prompt to an LLM on Amazon Bedrock, and posts a sho…
Tencent released Hy4, its first multimodal language model that understands images and can operate software on its own, claiming second place among Chinese AI models. Hy4 scores 35 points on the Artificial Analysis Intell…
Mistral AI expanded Mistral Studio to link its open-weight flagship model, Le Chonk, to managed agents, workflows, apps, enterprise context, observability and evaluation features that enterprises can run in their own clo…
NVIDIA says its full-stack codesign approach lifted agentic throughput per dollar by up to 30x from the Blackwell Ultra to the Vera Rubin platform while HBM capacity per GPU rose only about 3%, measured on SemiAnalysis A…
NVIDIA says its Vera Rubin NVL72 delivers up to 30x the throughput per megawatt of GB300 NVL72 on SemiAnalysis AgentX at 160 tokens per second per user, even though per-GPU HBM capacity held at 288 GB while bandwidth ros…
A practitioner blog post lays out three architecture patterns — orchestrator-worker, blackboard, and pipeline — for coordinating multiple AI agents, arguing that coordination, not reasoning, absorbs most of the engineeri…
AI interpretability startup Goodfire unveiled a technique on Nov. 5 that uses internal model signals to reward AI systems for reaching answers the right way, not just for getting them right, sharply reducing reward hacki…
A developer's writeup explains why AI models degrade on the back half of long documents, attributing it to three separate causes: context-window limits that silently truncate input, the 'lost in the middle' effect where …
An audit of Moonshot's Kimi K3.1 Pro model card found that the 1.56 TB, 2.8-trillion-parameter open-weight release loses to Claude Opus 5.5 on 20 of the 39 benchmark rows where both have scores, winning 19. The reviewer …
A tutorial published with a five-minute video shows developers how to build a working AI "harness" in 26 lines of free, open-source Python using PraisonAI, the open-source agent framework from MervinPraison, to stop a mo…
A GitHub project called ia-da-sobrevivencia, published by developer sdalberto, packages a fully offline survival assistant that runs on Windows, Linux and macOS from a USB drive, combining local language models via llama…
A November 4, 2026 arXiv paper (q-fin.CP, 2611.06607) describes Autonomous Quant Researcher, an open-source LLM research loop for quantitative finance in which the model proposes strict declarative JSON changes and only …
A Kagi Assistant user reported that uploading a PDF, selecting Gemini 4 Astra, and requesting OCR-to-Markdown conversion failed, with the tool returning "Min document size not reached" and no transcribed content. The fai…
A hands-on test by Kwangju Shin (KJ) found that Qwen3.8 Flash running on an RTX 4090 via the open-source Strata inference engine decoded at about 126 tokens per second in English and 118 in Korean with IQ3_XXS weights, v…
Inngest co-founder and CEO Tony Holdstock-Brown reported that server timeouts or dropped connections appeared in 41% of affected accounts running long-running agents, compared with 15% for LLM API errors, arguing the mod…
DeepSeek has opened the web search infrastructure behind its models to developers with two new endpoints, a Search API that returns up to 50 ranked results per response and a Fetch API that returns a public page's main c…
OpenAI's textGrain watermark, released in October to comply with the EU's new requirement that AI-generated text be identifiable, detected watermarks in about 80 percent of 200-token passages versus 95 percent of 400-tok…
Zep announced Zep Agent Skills, a production service that compiles an agent's verified runs into versioned SKILL.md procedures in the open Agent Skills format, with each step linked back to the run events it came from. I…