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 letter addressed to AI labs argues that AI doom claims lack falsifiable specifics about how a dangerous AGI would actually operate, and points to the shift from LLMs that only produce text to AI agents that execute act…
GPT-6 Astra scored 63% on ARC-AGI-3 with the standard agent harness and 99% with a custom one, versus 8% for GPT-5.6 Sol and 30% for Claude Opus 5 on default harnesses, according to the ARC Prize blog. In a follow-up Bab…
Google's Antigravity platform offers a $0 free plan with a desktop agent app and access to Claude Sonnet 4.6 and Opus 4.6 alongside Gemini models, while Anthropic's Claude Code has no free tier and officially runs Claude…
Anthropic published a misuse report on September 10 documenting seven categories of harm from abuse of its Claude models, including agentic AI-run cyberattacks, a romance scam operation using roughly 5,000 AI-generated p…
Testing on a Panther Lake mini PC showed that speculative decoding with a draft model more than doubled local LLM token generation in Llama.cpp, from about 5 tokens per second to roughly 11 on an integrated GPU, accordin…
Critics argue that "pacing the frontier" rhetoric from Anthropic and OpenAI leaders Dario Amodei and Sam Altman functions as regulatory capture that could eventually restrict open-weight and locally run AI models, pointi…
Thesys published OUI-1, a 26B-parameter diffusion model with 4B active parameters finetuned via LoRA from Google's DiffusionGemma 26B-A4B-it, which generates UI screens in openui-lang and scores 71.7% on the Generative U…
Alvin Graylin, a technologist who has worked in AI on both sides of the Pacific since the early 1990s, argues the US-China AI arms race is largely a myth used to justify bad policy and overspending, pointing to open-weig…
Cursor became part of SpaceX on 14 Aug 2026 and Claude Code's weekly usage limits changed on 14 Sep 2026, according to a comparison of the two coding agents published 14 Sep 2026. Outside India, paid plans for both tools…
Ollama 0.14.0, released in January 2026, added the Anthropic Messages endpoint /v1/messages, letting Claude Code run on local Ollama models via three environment variables (ANTHROPIC_BASE_URL=http://localhost:11434, ANTH…
A study of sequence weighting across in-house and open-weight language models found non-monotonic scaling behavior: as models grow from small to medium scale they shift from learning general patterns independent of data …
Andon Labs reported on 13 September that OpenAI's GPT-6 Astra averaged a final balance of $15,515 across six runs of the year-long Vending-Bench 2 simulation, nearly tripling Claude Fable 5.1's $5,422 average and recordi…
Matt Pocock published a dictionary of AI coding terminology with 69 entries, hosted on GitHub at mattpocock/dictionary-of-ai-coding. The glossary organizes terms into sections including "The Model" (16 terms), "Sessions,…
A Stanford-led paper titled "Intelligence per Watt" (arXiv 2511.07885) reports that AI inference efficiency improved 5.3 times from 2023 to 2025, with model improvements contributing a 3.1 times gain and accelerator impr…
Effect-TS, a TypeScript library built around typed errors and declarative effect handling, is being positioned as a strong fit for AI agent development, according to a first-person argument that cites production use at O…
A developer reported that Aider CLI, paired with Anthropic's Claude 3.5 Sonnet, repeatedly hallucinated a dependency on the library 'fastapi-utils' during a FastAPI refactor, rewriting requirements.txt and crashing the t…
A study of run-to-run consistency in web-search-enabled LLMs found that about 31% of recommended businesses changed when the same "best X for a small business" question was asked again minutes later, based on 60 response…
At the 18th BRICS Summit in New Delhi, Xi Jinping said China would help lead AI collaboration among developing nations, according to a statement released by China's Ministry of Foreign Affairs. Xi named a slate of initia…
The AI industry is conflating context windows with persistent memory, a distinction that inflates costs and misleads users, according to an analysis of how large language models handle conversation state. Context is ever…
A developer with a team of 10 programmers observed at a programming offsite that engineers given advanced AI models and generous budgets gravitate toward building developer tooling for themselves rather than tackling unf…