Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Chinese AI lab Zhipu AI, under its global brand Z.AI, launched the open-weight GLM-5.2 model with up to 753 billion parameters, rivaling top closed models from Anthropic and OpenAI at roughly one-tenth the cost. The mode…
KRAFTON built PUBG Ally, a co-playable character powered by NVIDIA ACE, for PUBG: BATTLEGROUNDS. The AI teammate uses automatic speech recognition, a 2B-parameter small language model, and text-to-speech to understand pl…
Tejas S Athni reported in the Journal of Medical Internet Research that conversational medical AI systems are increasingly simulating licensed health care providers while disclaiming legal responsibility, creating a gray…
A developer argues that maximizing AI context windows leads to worse results, advocating for optimizing relevance over volume. They found that flooding context with excessive information creates noise, causing slower res…
Prefill/decode disaggregation separates the two phases of LLM inference—prefill (compute-bound) and decode (memory-bound)—onto different GPUs to avoid the performance compromise of running both on the same hardware. The …
Multiplayer's debugging agent failed because raw observability data overwhelmed the AI, leading to incorrect fixes. The company developed a four-stage data curation process—grouping, correlating, deduplicating, and struc…
A new research paper reframes prompt injection as role confusion, arguing that dozens of named attacks are a single bug: models attribute authority by text style rather than structural role tags. The paper demonstrates t…
Ohad Ravid published an essay arguing that large language models impose social exhaustion by requiring users to expend social energy typically reserved for human interaction. The piece contrasts embodied tool use with th…
Dell Technologies introduced the PowerEdge XE8812 server at ISC 2026, a rack-scale platform built around NVIDIA's Vera Rubin NVL4 architecture that supports up to 144 GPUs per rack with over 300 kW power delivery. The fa…
The US Commerce Department issued an export control order on June 12 restricting foreign nationals from accessing Anthropic's advanced AI models, Fable 5 and Mythos 5, citing national security concerns. Anthropic disable…
A developer successfully ran Llama 3 locally using Docker and Ollama, achieving 2–4 second response latency on the 8B model. The setup provides privacy, full control over inference parameters, and offline availability, r…
A researcher ran a 200-step debug comparison between a base OLMo3 600M model and a DeepSeek-style Engram memory graft variant, finding the graft stable and showing improved early learning behavior. The Engram variant add…
Anthropic faced backlash after pulling its Claude Fable 5 and Mythos 5 AI models on June 12, following a US export control directive citing national security concerns. The company disabled access for all users, including…
Researchers at the Allen Institute for AI compared their 7B transformer model Olmo 3 with the hybrid model Olmo Hybrid to determine which tokens each predicts better. The hybrid model excels on meaningful tokens like nou…
A developer built a production-ready AI Chatbot SaaS platform using Next.js, TypeScript, and OpenAI. The application features user authentication, conversation management, and a ChatGPT-style interface with PostgreSQL da…
Researchers at Goodfire demonstrated that fine-tuning a single scalar prefactor on a German-related rank-1 parameter subcomponent of a 67M-parameter language model can destroy its ability to predict German text while pre…
OpenKnowledge, an open-source, AI-first markdown editor and LLM wiki, launched as an alternative to Obsidian and Notion. The tool offers WYSIWYG editing, collaborative AI-editing with Claude and Codex, and team sharing v…
A Washington Post investigation found that major AI chatbots, including OpenAI's GPT-5.5 and Elon Musk's Grok, predominantly present left-leaning political arguments, with Google's Gemini 3.1 Pro being the only exception…
A developer is building an AI Chatbot SaaS to strengthen full-stack engineering skills, using Next.js, TypeScript, PostgreSQL, Prisma, Tailwind CSS, and OpenAI API. The project includes user authentication, AI-powered co…
A small, energy-efficient AI model (3B parameters) has matched the performance of the much larger Opus 4.5 model in mathematics and coding tasks, challenging assumptions about the necessity of massive compute resources.