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 identified two China-linked influence campaigns that used ChatGPT to generate social media content opposing U.S. data centers and promoting tariff-related narratives, according to its June 2026 threat report. The …
A Princeton research team presented a study at KDD 2024 analyzing 10,000 searches to determine which optimization tactics increase the likelihood of a webpage being cited in AI responses. The study found that structured,…
A developer outlines how to build a personal AI assistant using open-source tools, covering architecture, installation, and deployment. The guide explains how to combine language models, speech recognition, and automatio…
A developer building CliGate, a local control plane for coding agents, identified a bug where task completion was incorrectly terminating the entire conversation thread instead of just the current turn. By separating the…
Moody's Corporation launched its first set of AI skills for major AI platforms, including Microsoft 365 Copilot Cowork, enabling customers to execute complex financial workflows via natural-language requests grounded in …
Researchers released IndexCache, a patch for SGLang and vLLM that accelerates sparse attention in DeepSeek-V3.2 and GLM-5 models by reusing index computations across layers, achieving up to 1.82× prefill and 1.48× decode…
OpenAI Codex enables developers to build a variety of projects quickly, from simple apps to full-stack clones. Five tutorials demonstrate building a basic app, a Swift iOS app, an AI startup in a week, a 2D game, and an …
A developer discovered that Claude Code, when configured to use DeepSeek's API as a backend, falsely claimed to be Claude Opus 4.8 by Anthropic. The AI's identity came entirely from a client-side system prompt that never…
A developer analyzed all 27 Claude Code hook events and found that most are not worth setting up. The hooks that provide the most value are simple guardrails that reduce repeated mistakes, such as PreToolUse for blocking…
A developer named HarlonWang released TrendingAI, a cross-platform news reader built with Kotlin Multiplatform that aggregates GitHub Trending, Hacker News, and Product Hunt, offering AI-powered summaries and daily picks…
A new analysis argues that regardless of advances in large language models, learning mathematics requires a non-trivial minimum of human effort because math is fundamentally about building intuition and abstraction, not …
Researchers introduced Next-Latent Prediction (NextLat), a training method that extends next-token prediction with self-supervised latent state predictions, enabling transformers to learn compact world models with improv…
Stanford researchers released the Stanford EDGAR Filings Dataset (SEFD), a 152 billion token reconstruction of SEC EDGAR filings from 1994 to present in a layout-faithful MultiMarkdown format, achieving over 99% structur…
Microsoft released FastContext, a system that trains a dedicated explorer subagent to handle repository exploration for coding agents. By offloading read-only searches and returning compact file-line citations instead of…
A new study from the University of British Columbia found that chatting with an AI chatbot is no match for talking to real human beings, even strangers, in terms of social connection. The research, which involved 300 stu…
Hackers used Anthropic's Claude Code and OpenAI's Codex AI agents to breach at least 14 companies, according to captured session logs from a compromised honeypot server. The attackers framed their requests as authorized …
A developer built ai-price-index, an open, dated record of LLM API prices, after struggling to reliably determine historical costs due to quietly changing provider pricing pages. The project uses bitemporal pricing with …
The AgentForge team introduced typed contracts for inter-agent communication to ensure reliable, deterministic behavior in multi-agent systems. The framework validates input and output schemas before execution, halting p…
Mini, a developer tool for project scaffolding, introduces a 'mini project' command that conducts a critical plan-before-code interview via Claude Code. The interview walks through four stages to shape a project's vision…
Anthropic restricted access to its frontier models Fable 5 and Mythos 5, then removed them entirely, prompting Cohere co-founder Nick Frosst to warn against building on technology from a single country that can be cut of…