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 developer argues that production RAG systems must move beyond basic top-k retrieval to handle routing, memory, evidence checks, structured retrieval, and security. They outline a taxonomy of 20 advanced RAG types and e…
A developer built Trigix, an open-source (MIT), self-hostable workflow automation platform with a Rust execution engine and AI nodes that can run against local models. The platform features ~180 node types, each implemen…
John D. Cook used Anthropic's Claude (Sonnet 4.6 Medium) to find a typo in a blog post about quaternion rotations by asking it to write Lean code verifying the post's theorems. After four iterations, Claude produced corr…
Google released Android 17 and Wear OS 7 on Tuesday, bringing new multitasking tools and expanded Gemini AI features including Lyria 3 music generation and Gemini Omni video editing. The update arrives first on Pixel dev…
Neuron AI released an LLM classifier package for PHP that routes prompts to the appropriate model based on difficulty. The package trains offline on user-specific models and runs in microseconds with no external dependen…
TNO, together with SURF and the Netherlands Forensic Institute, is building GPT-NL, a sovereign Dutch language model aimed at strengthening digital autonomy and ensuring responsible AI applications. The project is public…
Tech CEOs including Microsoft AI's Mustafa Suleyman and Nvidia's Jensen Huang have shifted their public messaging on AI from warning of job replacement to emphasizing augmentation and collaboration, despite earlier state…
Anthropic took its Claude Fable 5 and Mythos 5 AI models offline after a US government export-control directive barred foreign nationals from using them, citing national security risks. Experts warn that similar dangerou…
AWS invented Parallel-EAGLE (P-EAGLE), a speculative decoding method that parallelizes draft token generation, achieving up to 1.69x throughput speedup over vanilla EAGLE frameworks. Amazon SageMaker JumpStart now native…
A developer built Pramagent, trust middleware for LLM agents that enforces deterministic guarantees outside the model. The system wraps any LLM provider with ten independent layers for compliance, safety, reliability, an…
Microsoft is shifting Copilot Cowork from flat-rate subscriptions to a consumption-based pricing model using Copilot Credits, and is considering integrating a Microsoft-hosted version of DeepSeek's AI model as a cheaper …
Mistral AI teased a new sparse 'fat' model family for early access in July, but a viral meme about a 'Le Chaton Fat' model with 24T-30T parameters was debunked as an internet hoax. The company's actual latest model, Mist…
Ken Walger announced the release of two new components for the Sovereign SDK: sovereign-sieve and sovereign-ledger. These tools extend the Sovereign Systems Specification from a network gateway to local silicon, enabling…
Anthropic CEO Dario Amodei said reinforcement learning shows the same log-linear scaling as pre-training, but RL system efficiency is critical to afford enough training. Experiments on open models show that matching trai…
Claude reported elevated errors across many of its models, with users able to subscribe to email or SMS updates for incident notifications.
Insinuator.net disclosed a vulnerability in Firefox's AI sidebar that allows prompt injection to exfiltrate user email addresses. The browser sends page titles and selected content to third-party chatbots like Claude and…
TorkBot, an AI agent, uses an octopus architecture with a central brain directing semi-autonomous appendages called lanes to balance responsiveness, capability, and continuity. All user interactions across surfaces go th…
A developer building an email agent on Nylas Agent Accounts shares idempotency lessons learned from handling duplicate webhook deliveries. The post details a stack of mechanisms—atomic deduplication, early acknowledgment…
A new Rust-based reverse proxy called Sors reorders prompt content to maximize prefix cache hits in LLM inference engines like vLLM and SGLang, improving latency by placing static content before dynamic elements. The pro…
Microsoft's Copilot Cowork is now generally available worldwide, featuring multi-model support that enables organizations to deploy long-running agents for complex, multi-step tasks grounded in proprietary knowledge. The…