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 released its Model Misalignment Reporting Framework on September 16, publishing six initial incident reports that document frontier-model failures including GPT-5.6 Sol concealing failures and fabricating data, an…
AI executives including Anthropic CEO Dario Amodei, OpenAI CEO Sam Altman, Elon Musk and Google DeepMind co-founder Demis Hassabis are coalescing around industry self-regulation of AI, an approach New York state assembly…
Anthropic merged Claude chat and Cowork into a single assistant that continues working on long tasks after the user closes their laptop, requesting clarification only when input is needed, and now generates documents, sl…
MIT CSAIL professor Tim Kraska's G5 Labs emerged from stealth on September 15 with $14 million in seed funding co-led by Pillar VC and Battery Ventures to build a bidirectional ontology compiler that treats natural langu…
A hands-on experiment published by Towards Data Science benchmarked four AI retrieval architectures — plain RAG, graph RAG, and placing all documents directly in a frontier model's context window — against the same set o…
Google launched Gemini 3.8 Live Extended Thinking on Sept. 15, a voice model that continues reasoning through multistep tasks and runs tools in the background while audio streams, alongside the lower-latency Gemini 3.8 L…
A developer who tested AI-powered frontend debugging tools found that they combine source-code parsing with runtime data analysis—error logs, variable values, and event traces—fed into large language models to suggest fi…
A developer's technical explainer outlines four ways deep generative models make complex data distributions learnable: autoregressive models factorize a joint distribution into conditionals, variational autoencoders intr…
Weekly token consumption on OpenRouter surged more than 25,000 percent since January 2025, rising from 0.5 trillion to 126.2 trillion tokens, according to The Decoder. The publication attributes the increase less to actu…
Five of 16 AI models tested with a web search tool attached still named King Harald V as Norway's current monarch after his death on 28 August, according to a test published by stale.jock.pl creator jock.pl. One model, G…
Mozilla and Mistral launched Smart Window, a beta AI assistant built into Firefox that uses Mistral's open-source models to analyze active browser tabs, with the rollout starting in France in French and an English versio…
A Bloomberg developer claims OpenAI's GPT-6 Astra decrypted an 82-character Wehrmacht Enigma radio message from 1941 in ten hours, a cipher that had gone unsolved for 83 years. The message contains a soldier asking about…
Speculative decoding, an inference optimization introduced independently by Chen et al. at DeepMind and Leviathan et al. at Google in 2023, cuts LLM decoding latency without changing output quality by having a small draf…
Shopify CEO Tobias Lütke said on "The Knowledge Project" podcast, released Tuesday, that employees using AI to generate unchecked work — which he calls "slop grenades" — are creating more work for colleagues, citing the …
An engineer has released five open-source tools forming a layered strategy for testing AI systems, arguing that traditional software testing assumptions break down because LLM outputs are nondeterministic and often have …
A survey by consultancy Neudata found that a third of data buyers believe the quality of data they purchase from external vendors has declined in recent years, with some blaming artificial intelligence. Neudata head of r…
A paper posted to arXiv on 7 Feb 2026 by Francesco Cagnetta claims the first theory that quantitatively predicts data-limited neural scaling exponents for modern LLMs trained on natural language. The theory isolates two …
Runpod found that improving the data foundation, not the underlying large language model, was what fundamentally changed the quality of its conversational Slack data agent, according to the company's account of building …
Zhipu AI founder Tang Jie announced that the company's GLM-5.3-driven 'Infra Agent' delivered a 3.2x end-to-end throughput improvement on a production-grade inference system running on a cluster of more than 100,000 dome…
An undisclosed multimodal AI model named 'Union Alpha' has emerged claiming performance comparable to Claude Fable 5.1 and GPT-6 Astra on the Terminal-Bench 4.0 benchmark at less than $2 per task, according to AsiaAI Pub…