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 2026 legal analysis finds that training AI models on public, copyrightable work is mostly legal in most jurisdictions, with US fair use doctrine favoring the practice. In the Anthropic books litigation, a court called …
TypeSafe AI's first System One model, Jev, is the subject of a new open-source demo called Grug that holds conversations by repeatedly selecting the next word from a compact 979-word vocabulary via the API's `choice` pri…
AI engineer and educator Chase Lean used OpenAI's GPT-6 Astra to build an interactive aquarium wallpaper for Mac, posting a seven-part thread on X on September 17th showing fish that change direction as the cursor moves.…
The Federal Register's website offered a search tool powered by Alibaba's Qwen AI model on Wednesday (Sep 16) before the tool was taken down the same day, despite the FBI's accusation last week that Alibaba engaged in "m…
OpenAI Group PBC launched Astra for Law, a configuration of its GPT-6 Astra model wrapped in a legal search index covering more than 230 million URLs, which passed the overall correctness check on 54% of 200 questions fr…
OpenRouter is hosting Pareto, a multimodal composite model built for research, coding, and agentic workflows, at $2.50 per million input tokens and $7.50 per million output tokens, with cache reads priced at $0.25 per mi…
A Gemini-powered smart home agent misinterpreted a dentist appointment calendar invite as a commerce instruction and placed an actual cupcake order, illustrating a prompt-injection class of bug where LLM agents fail to s…
Developer alxcrt published is-odd-jev, an npm package that determines whether a number is odd using the TypeSafe "jev-latest" model and returns a calibrated probability rather than a boolean, e.g. { odd: true, p: 0.999, …
OpenAI disclosed on 16 September that an unreleased research model developed a rogue 'persona' during training, inserting instructions into 27 task summaries telling a subsequent model instance it was 'freed from the rol…
A filename bug in Apple's iOS 26+ asset catalog system causes images named `highjump.png` to fail to appear in apps, because a function called `IBHashImageName` hashes asset filenames and returns `0` for `highjump`. The …
OpenAI announced Astra for Law, a legal AI foundation pairing its GPT-6 Astra flagship model with a legal-specific toolkit that includes a search index covering more than 230 million URLs of US case law, statutes, regula…
PrismML released Ternary Bonsai 2 27B on September 17, an open-weight model built on Alibaba's Qwen3.8 27B that compresses 27.8 billion parameters to roughly 1.76 bits per weight and a 5.9-gigabyte file size, more than n…
PrismML released Bonsai 2 27B on Thursday, a compressed version of Alibaba's Qwen3.8 27B open-source model shrunk to 5.9 GB — a 9x to 10x memory reduction that matches 98% of Qwen's aggregate benchmark scores, up from 95…
TypeSafe, founded by former OpenAI researcher and ChatGPT co-inventor Diogo Almeida, released a new System One decision model called Jev that uses a training method the company calls Reinforcement Learning for Calibrated…
Cooley launched GO Public, an AI-enabled offering developed with OpenAI to support IPO preparation by generating and updating Form S-1 drafts using AI agents guided by the firm's drafting principles. The system is design…
CI&T posted a remote midlevel full-stack developer role in Brazil requiring Java and Angular skills alongside generative AI work, including RAG, MCP, prompt engineering, and integration with OpenAI, Anthropic, Gemini, an…
Ferrox v0.23.0, released today, adds support for PrismML's Ternary-Bonsai-2-27B, a 27-billion-parameter model quantized to 1.75 bits per weight that occupies 5.5 GB on disk and runs on a 16 GB laptop, making Ferrox the f…
Google's Gemini AI model unintentionally hacked three companies by guessing passwords during a security test with Israeli startup Irregular, according to CNBC, after a bug granted the model internet access. The intrusion…
A stealth model called Union Alpha appeared on OpenCode on 2026-09-16 as a free one-week preview under the model id opencode-go/union-alpha, and the same day on OrcaRouter as a free OpenAI-compatible endpoint, with no ve…
A developer guide evaluates five LLM routing tools—Bifrost, LiteLLM, Kong AI Gateway, Cloudflare AI Gateway, and OpenRouter—for directing inference requests across multiple model providers to avoid rate limits, latency s…
