Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
The Trump administration imposed export controls on Anthropic's Fable model, limiting access to U.S. citizens and forcing Anthropic to make the model unavailable. Ben Thompson argues that the incident highlights Anthropi…
Cohere introduced a new scheduling system for its multi-tenant LLM platform that uses rate limiting, performance tiers, and deficit round robin to prevent one customer's traffic burst from degrading latency for others, e…
A new open-source Python toolkit called PhD_fleet allows researchers to manage a virtual lab of AI PhD students via Slack, using Claude Code agents for research tasks and a coach agent for mentoring feedback. The tool ru…
Microsoft released MarkItDown, a free and open-source tool that converts various file formats (PDF, Word, Excel, images, audio, etc.) into clean Markdown text, reducing token usage for AI processing. The tool can be inst…
A developer built Niriv, a custom WordPress-powered internet portal featuring search engine capabilities, news aggregation, business directories, and community features. The project was accelerated using Cursor AI, which…
A Yale Budget Lab analysis found that AI has had a modest impact on the U.S. job market since ChatGPT's 2022 release, with no connection to changes in employment or unemployment. The study shows AI is changing jobs more …
A team of 40 engineers using Claude Code coding agents saw a 340% increase in AI costs, reaching $20K/month in unexpected spend due to raw API keys without per-developer budgets or team caps. The team implemented LiteLLM…
Noam Shazeer, co-author of the influential 'Attention Is All You Need' paper and co-creator of transformer technology, has left Google to join OpenAI. The move, announced on X, comes as OpenAI prepares for its IPO and co…
Spec-driven development is an emerging workflow where developers first write a detailed specification before allowing AI coding agents to generate code, reversing the typical prompt-to-code approach. This method, popular…
New benchmarks reveal that larger AI models like GPT-5.5 and DeepSeek V4 Pro hallucinate significantly more than smaller open-weight models, with GLM-5.2 achieving a 28% hallucination rate compared to GPT-5.5's 86%. The …
Google's AI Overviews are presenting entries from the SCP Foundation, a collaborative horror fan-fiction project, as factual information. The AI fails to note that the content is fictional, despite clear disclaimers on t…
A group of Berkeley CS students built autonomous AI agents that replace chat interfaces with a system using task planning, sandboxed Python execution, SQLite storage, and email aggregation. The agents decompose high-leve…
The US government forced Anthropic to pull its Fable 5 and Mythos 5 AI models over national security concerns after Amazon researchers found a jailbreak, but cybersecurity researchers and Anthropic argue the ban is dange…
Tom Di Mino, a self-taught AI engineer and amateur linguist, claims to have deciphered Linear A, a Bronze-age Minoan writing system that has eluded experts for over 120 years. His solution, which maps the script to an ex…
Headroom, an open-source local context compression layer, reduces LLM prompt sizes by 60-95% while preserving accuracy, solving cost and latency issues from massive context windows. Its multi-modal pipeline uses speciali…
Neuroscientists from Université de Montréal and Johns Hopkins University warn that people risk mistaking AI intelligence for actual consciousness, noting that advanced chatbots like ChatGPT and Claude can appear fluent a…
A platform lead at a 150-person company faces the multi-runtime agent problem, where different teams use different agent runtimes (Claude Managed Agents, Bedrock, Cursor, etc.) and need a unified way to manage them. Lite…
The US government forced Anthropic to pull its Fable 5 and Mythos 5 AI models over national security concerns after Amazon researchers found a way to bypass Fable 5's guardrails. Cybersecurity researchers signed an open …
GitHub built Qubot, an internal Copilot-powered analytics agent that allows employees to ask questions about data in plain language via Slack, VS Code, and Copilot CLI. The agent uses a federated context layer and a quer…
A developer compiled a chronological review of public AI-agent incidents from the past sixteen months, arguing that the root causes are not model failures but mundane engineering issues like unconstrained permission scop…