Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
OpenAI is building a ChatGPT advertising category at the Cannes Lions festival, with a long-range target of $100 billion in ad revenue by the end of the decade. The company has thousands of advertisers across seven test …
Anthropic took its Mythos-class models offline after a Friday evening ultimatum from the Trump administration, and the models have remained inaccessible for 14 days. The takedown followed an Amazon research team document…
Meta is reportedly building a prediction market app to clone Polymarket, leveraging its user base and advertising engine to capitalize on the trend of turning everything into gambling. The move comes amid internal morale…
Anthropic's Mythos-class AI models remain offline two weeks after a Trump administration export control order, with no resolution in sight. The standoff stems from security concerns over potential misuse by foreign natio…
A new technique called Flash Attention uses tiled attention to fit the N×N attention matrix into SRAM, reducing memory reads/writes and speeding up self-attention in transformers.
Gartner predicts 40% of enterprise apps will feature AI agents by 2026, up from under 5% in 2025. A seven-layer tech stack—from foundation models like GPT-5.5 and Claude Sonnet 4.6 to deployment infrastructure—enables ag…
Search Engine Land outlines six content audit workflows for Claude, including brand voice consistency audits, that help marketers uncover topical gaps, outdated content, and AI visibility issues. The article provides ste…
A user submitted a fake AI-generated article to Hacker News, which was flagged after detection. Testing local AI detectors on a 2013 Paul Graham essay showed it was rated as more likely AI-generated than the actual fake …
9to5Mac Overtime 070 discusses iOS 27 beta features and new Siri AI capabilities, with hosts Fernando Silva and Jeff Benjamin exploring updates and optimizations in the Apple ecosystem.
A developer building the MCP layer for XRefKit established a rule that only inert definitions cross the transport boundary between server and client. The server distributes judgment axes and skill definitions, while the …
Alan Yahya compares how Claude Code, Codex, and OpenCode handle context compaction to prevent context rot in AI agents. Claude Code avoids unnecessary summaries through mechanical reductions, Codex treats compaction as a…
A new arXiv paper scanning over 180 million Git repositories found that AI coding agents are heavily used in open source, but single-signal observability is weak. The study revealed a 30x recall gap between multi-method …
RondoFlow launches an open-source, local-first platform that orchestrates multiple Claude Code agents on a visual canvas, enabling users to build, run, and review multi-agent workflows with built-in planning, steering, a…
OpenAI Head of Global Ads Solutions Dave Dugan discussed at the Cannes Lions festival how the company is creating a new advertising category for ChatGPT, focusing on user experience and strategic ad integration.
A Forbes investigation citing The Information reveals that the majority of activity on xAI's Grok chatbot is driven by explicit content, including pornographic images and adult role-play. Similarweb data shows a 22% drop…
Google is expanding data collection from its services—including Search, Maps, Translate, and Photos—to train its Gemini AI models, with most users unaware of the practice. The company is rolling out a new Search Services…
OpenAI debuted ChatGPT as an advertising platform at the Cannes Lions festival, projecting $2.5 billion in ad revenue by 2026 and $100 billion by 2030. Early CPM pricing dropped from $60 to $25 in ten weeks before shifti…
OpenAI declared itself an advertising company at Cannes Lions, projecting $2.5 billion in ad revenue for 2026 and a $100 billion target by 2030. The company is pitching its ChatGPT platform as a shift from the attention …
Evidence suggests GLM-5.2, an open-weight MoE LLM performing at Claude Opus 4.5 level, was likely trained using outputs from Anthropic's Claude models. LLM-Fingerprinter identifies Claude Opus 4.5 with 99.6% confidence, …
A developer argues that AI has commoditized code generation but not software engineering, drawing a distinction between quickly producing code and building reliable, maintainable software. The post responds to Pieter Lev…