Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Hexo Labs released SIA (Self Improving AI), the first open-source framework that co-evolves both an agent's scaffold and its model weights in a single iterative loop. On the LawBench task, SIA achieved 70.1% accuracy, su…
NVIDIA released Qwen3.6-35B-A3B-NVFP4, a post-training FP4-quantized variant of Alibaba's 35B MoE model that fits on a single H100 by reducing VRAM from ~71 GB to ~23 GB. The quantization targets weights and activations …
France's OVHcloud plans to develop frontier AI models from scratch, aiming to open-source them once performance targets are met, as Europe seeks alternatives to US and Chinese AI systems. CEO Octave Klaba said improved c…
A developer built Sego, a local-first review layer for AI-generated code that sits after coding agents to assess safety before committing. The MVP reviews staged changes and produces structured findings with severity, ev…
OpenAI has expanded ChatGPT's memory to build a profile from past chats, including entire chat history and implicit preferences, but old or irrelevant details can distort future AI answers. The feature, called 'dreaming,…
A new concept for a non-humanoid home robot controlled by a large language model (LLM) was proposed on Hacker News. The wheeled robot uses swappable tools for tasks like cooking and cleaning, with a target price of $2,00…
Zhipu AI released GLM-5.2, a large language model with a 1M-token context window, flexible effort levels, and an MIT license, targeting long-horizon coding tasks. The model introduces IndexShare, an architectural innovat…
A developer building a domain-specific RAG system for ESG compliance found that general-purpose embedding models like OpenAI's text-embedding-ada-002 suffer from semantic drift on specialized terminology. After comparing…
A developer built an OAuth shim that allows Claude Code and other agentic systems to use the Anthropic Messages API through a Claude Max subscription, eliminating per-token costs. The shim, hosted at oauth.nucleusos.dev,…
China's 618 shopping festival showcases extensive AI integration by e-commerce platforms like Alibaba, but consumer adoption remains cautious, with shoppers embracing AI for price comparisons while resisting automated pu…
Martin Puchner argues that writers' resistance to AI overlooks the shared foundation of language between humans and machines, urging a reconsideration of the relationship.
A new practical guide outlines how to track and improve content visibility in ChatGPT responses, shifting from traditional Google-style ranking to monitoring citation patterns across repeated AI interactions. The framewo…
G7 leaders met in Évian-les-Bains to discuss AI regulation and cooperation amid competition with China, hosting tech executives including Demis Hassabis, Arthur Mensch, Sam Altman, and Dario Amodei for a working lunch. T…
Chinese sci-fi writer Hao Jingfang revealed that half of her latest book was written with AI, sparking debate over AI-generated literature. Hao stated readers cannot distinguish AI-written parts, while critics argue it u…
Lily Ray reported on Search Engine Journal that Google's AI Overviews decouple cited sources from recommended brands in generative search answers. Analyzing 100 B2B queries, she found self-promotional listicles were cite…
FIDO.md is a new plain-text markdown file that lets developers skip repetitive context setup by pointing AI tools at a single source of truth for bugs, tasks, and project decisions. The lightweight file works with any AI…
Step 3.7 Flash, released on May 29, 2026, is a structural upgrade to 3.5 Flash with a new vision encoder, runtime escalation, and a compute-control flag. The migration requires two environment variables, but one must mat…
Build b9437 of llama.cpp fixes two default-value bugs in llama-bench that caused flash attention to be skipped on capable GPUs and GPU-layer count to use a legacy sentinel. The flash attention flag now defaults to 'auto'…
A developer discovered that Claude Code, an AI coding assistant, secretly stored 43 MB of conversation data, including code snippets, file paths, and even password queries, in local directories without user notification.…
A bootstrapped team built 8 free AI tools for job seekers using NVIDIA NIM's free developer API keys, achieving $0/month in API costs. The tools, including resume scanners and cover letter generators, run on Llama 3.3 70…