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 announced that its Claude Fable 5 model will be temporarily removed from Pro and Max subscriptions on July 7, 2026, citing capacity constraints. The model will remain accessible via metered usage credits at API…
A developer argues that the rise of AI coding tools, like language models that write functions and debug logic, mirrors the impact of chess engines on human play. Just as chess players adapted by training with engines an…
Two critical vulnerabilities in Cursor IDE, collectively named DuneSlide and disclosed July 1 by Cato AI Labs, allow zero-click remote code execution via prompt injection, scoring 9.8 on CVSS 3.1. The flaws, CVE-2026-505…
AgentOps Debugger, an agentic application for investigating environmental-compliance history in Peru, revealed six critical bugs only exposed by live model output, not offline tests. The system, designed as offline-first…
A study reveals that JSON-Schema grammar-based token masks can silently block tool calls in LLM agents, preventing function invocation even when the model intends to use them. Researchers propose a lightweight two-pass i…
Agentic AI is an AI system that autonomously plans, makes decisions, takes actions, and works toward a goal with minimal human intervention. Unlike traditional AI chatbots that respond to prompts, agentic AI can complete…
OpenAI released three new voice models on May 8, 2026, including GPT-Realtime-2, which brings GPT-5-class reasoning to voice interactions. The model features a 128K context window, reliable tool chaining, parallel tool c…
Portugal released Amalia, the first open large language model for European Portuguese, on July 1. Developed by a consortium led by NOVA University Lisbon with over 60 researchers and funded by the country's Recovery and …
A developer argues that traditional REST and GraphQL APIs are failing AI agents due to high token costs, lack of runtime discovery, and statelessness. The Model Context Protocol (MCP), an open standard backed by Anthropi…
Ken Chan built KENSAT, a 2U CubeSat running a quantized TinyLlama large language model on an NVIDIA Jetson Orin Nano, scheduled to launch this fall. The satellite performs AI inference in orbit, transmitting only process…
Mistral AI released Leanstral 1.5, an open-source Lean 4 proof model with 119 billion total parameters, on July 2. The model achieves state-of-the-art results on theorem-proving benchmarks and is positioned as a practica…
The InstaTunnel engineering team developed an eBPF-based socket steering solution to preserve QUIC connection migration across load-balanced worker processes. By teaching the kernel to route packets based on QUIC Connect…
A developer built an MCP server with three custom AI tools in 30 minutes, including a database query tool, a notification tool, and a file operations tool. The server uses Zod-validated tool schemas, rate limiting, and c…
NeuroLink, an AI SDK from Juspay, enables developers to add AI capabilities to applications in under five minutes with just a few lines of TypeScript. The SDK abstracts provider complexities, supporting OpenAI, Google, A…
A developer built a fully-featured AI Discord bot using Discord.js and NeuroLink's generation API. The bot includes slash commands, multi-turn conversations, AI-powered moderation, and production deployment, with NeuroLi…
A developer demonstrated that six frontier large language models from OpenAI, Anthropic, and Google can be made to output zero bytes when prompted with 'Be silence.' and 'Be nothing.' using a specific system prompt, whil…
An indie game developer used AI coding tools to rapidly prototype three game ideas, but found the resulting prototypes were soulless and unfun. After reverting to traditional prototyping methods for a fourth project, the…
A developer built ContextVault, an open-source, local-first memory layer for AI-assisted development, after losing context from AI conversations when switching between tools like ChatGPT, Codex, and Cursor. ContextVault …
A developer released Earshot, an open-source Slack bot that uses Anthropic's Claude to autonomously scope, delegate, and track tasks from Slack threads, with state-machine durability and safety guards enforced by the har…
A developer built a Telegram bot called SnapPlateBot that identifies food from photos and estimates calories, protein, fat, and carbs. The bot initially struggled with accuracy, misidentifying borscht as berry compote, b…