Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Anthropic warns that recursive self-improvement (RSI) in AI could arrive sooner than institutions are prepared for, as models like Claude now write over 80% of code merged into Anthropic's systems. The company calls for …
An engineering team deployed a market research pipeline using four LangChain agents in November 2025. A logic failure caused two agents to enter a recursive loop, resulting in a $47,000 API bill due to infinite hallucina…
Meta unveiled Brain2Qwerty v2, a non-invasive AI system that translates brain activity into text with 61% average word accuracy, compared to 8% for previous non-invasive methods. The system uses a magnetoencephalography …
A developer built higi, a self-healing structural middleware layer that sits between raw LLM strings and strict business logic to prevent production crashes caused by malformed JSON or other structural errors. Using a si…
Quesma engineer Piotr Migdal tested Alibaba's Qwen 3.6 27B model for local development and concluded it is viable for coding and general-purpose tasks, shifting the threshold for when developers should use local models o…
Arena, the UC Berkeley-born AI evaluation platform behind the Chatbot Arena leaderboard, tripled its annualized revenue from $30M to $100M in six months, but its business model of charging AI companies for evaluations ra…
Anthropic's Claude models are now generally available in Microsoft Foundry on Azure, offering Claude Opus 4.8 and Haiku 4.5 with features like prompt caching and extended thinking. The integration provides Azure-native a…
A developer merged Andrej Karpathy's CLAUDE.md and LOOPS.md with a battle-tested system prompt used in over 100 full-stack apps, creating a 35-rule, 6-tier framework for reliable AI agent behavior. The rules emphasize sc…
EthicalGuard, a new framework for AI safety, aims to help developers integrate robust governance and compliance into LLM-powered applications. The framework decouples core application logic from safety layers, addressing…
A quality manager with no programming background built a constitutional governance runtime for autonomous AI systems, called CORE, using only questions and AI assistance. The system enforces a strict separation between r…
California Governor Gavin Newsom and Anthropic have reached a deal allowing state agencies to use the Claude AI chatbot at a discounted price, aiming to improve government efficiency. The agreement diverges from the fede…
A developer explains why they refuse to run untrusted AI models in coding agents, citing risks of backdoored code and arbitrary execution. They trust only Anthropic and OpenAI due to legal and financial incentives, while…
Meta relied on Google's Gemini models for customer service, ad tools, and content moderation until Google imposed capacity limits in March 2026, forcing Meta to ration AI tokens and shift workloads to its own Muse Spark …
Amazon engineers are distilling Anthropic's AI models into smaller, cheaper versions for internal use ahead of a switch to token-based pricing next year, which could increase costs. The company is also exploring alternat…
Researchers from an undisclosed institution conducted a systematic study of LLM agent memory systems, evaluating 12 systems across 11 datasets. They found no single architecture dominates, with effectiveness depending on…
VLLM Semantic Router introduces a new open-source serving primitive that turns a single model API call into a bounded collaboration among multiple micro-agents, enabling cost savings, safety enforcement, and improved out…
Researchers introduced DiScoFormer, a transformer model that estimates both density and score of a distribution from a set of data points in a single forward pass without retraining. The model uses cross-attention and a …
A developer released Context Warp Drive, an open-source tool that deterministically folds LLM agent context without summarization calls, achieving ~90% cache-hit rates in production Claude workloads and reducing input co…
Air Canada was ordered to pay CAD $812.02 in damages after its chatbot provided false bereavement fare information, and attorney Steven Schwartz was fined $5,000 for submitting a legal brief with fabricated cases generat…
An open-source browser automation MCP called Brocogni reduces token costs by 81% compared to Playwright MCP by pre-computing selectors server-side. The developer found that iteration, not observation quality, drives cost…