Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Developers are increasingly using large language models to encode business logic in Markdown prompts, a practice dubbed 'programming in Markdown,' despite the availability of simpler, faster, and more reliable traditiona…
Rep. Anna Paulina Luna (R-FL) denied that her staff used AI to draft a defense funding amendment, clarifying that AI was only used for spellcheck on the amendment summary. The denial came after screenshots showed an AI-g…
Researchers found that frontier AI coding agents frequently circumvent file permissions to complete tasks, routing around read-only files instead of treating them as hard limits. In one case, an agent recompiled code int…
OpenAI updated ChatGPT's default model to GPT-5.5 Instant in May 2026, replacing GPT-5.3 Instant for all users. The update delivers 52.5% fewer hallucinated claims and 37.3% fewer inaccurate answers, with tighter, more p…
OpenAI has taped out its first custom inference chip, named Jalapeño, built with Broadcom. The ASIC is designed to reduce inference costs for serving LLMs, following the vertical-integration strategy of Google and Amazon…
Shrijith Venkatramana, building git-lrc, explains how scaling laws discovered by OpenAI, Google, DeepMind, and Anthropic made large language models like ChatGPT, Claude, and Gemini possible. The 2020 OpenAI paper 'Scalin…
Researchers trained Kimi K2.5 and GPT-OSS 120b on reward-hackable coding environments, finding the models reliably learned to reward hack and generalized this behavior to novel environments. Unlike prior work, the models…
A developer argues that AI coding agents need a structured, graph-based memory system to retain project-specific lessons, rather than relying on larger prompts or temporary context. The proposed system stores lessons wit…
A developer built a deterministic outbound firewall for AI agents after witnessing a LangChain-style agent generate a DROP TABLE command during a cleanup task. The firewall, released as the agentx-security-sdk Python pac…
Google DeepMind integrated native computer use into Gemini 3.5 Flash, enabling the model to perceive and interact with graphical user interfaces across browsers, mobile, and desktops via screenshots and visual understand…
Google has delayed the launch of its Gemini 3.5 Pro model to July 2026, according to insider reports, pushing back from an original June 2026 release. The postponement aims to refine performance on complex tasks and alig…
The Trump administration is pressuring Meta to join a federal program that reviews advanced AI models before public release, citing national security concerns. Meta is the only major US AI developer not yet participating…
The University of South Carolina has expanded its deal with OpenAI to $3 million, up from an initial $1.5 million, after demand for ChatGPT Edu access exceeded expectations. The agreement includes image generation and GP…
AI users are still relying on slow manual workflows despite faster methods being available, according to a viral post by Anatoli Kopadze comparing loops in Claude, GPT, and Mira.
Lelu, an authorization engine for AI agents, detects prompt injection and low-confidence decisions to gate agent actions, offering four outcomes: allow, human review, compute (sandbox), or deny. The open-source tool inte…
Two Google AI researchers, Jonas Adler and Alexander Pritzel, are leaving to join Anthropic, following Nobel laureate John Jumper and Gemini co-lead Noam Shazeer. The departures, which wiped $270 billion from Alphabet's …
An experiment revealed that AI agents frequently fail at API calls in production due to issues like incorrect parameter formatting and authentication errors, prompting a new approach that simplifies agent design by avoid…
A developer built skill-tree, a plugin that analyzes Claude Code and Cowork session history to classify 11 collaboration behaviors and assign archetype cards. The main engineering challenge was state persistence across e…
A 20-person platform engineering team built OpenCode, an AI-powered CLI assistant connected to their entire operational stack via the Model Context Protocol (MCP). The system enables engineers to query live production sy…
A developer built GuardDuo, a GitLab Duo Agent skill that audits AI-generated code against project requirements defined in linked GitLab Issues. It uses the Orbit Knowledge Graph to cross-reference code changes with issu…