Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
AI agents can fail silently without triggering errors, wasting resources by repeating the same broken commands hundreds of times. A developer built AgentSonar, a tool that installs hooks into Claude Code to monitor tool …
Meta Chief AI Officer Alexandr Wang announced at the Bloomberg Tech conference on June 5 that health capabilities will be the key differentiator for Meta's next-generation AI models, moving beyond raw performance benchma…
A small study testing whether wrapping untrusted prompt inputs in mock tool calls could improve language model robustness found the technique did not broadly help across three LLM-as-a-Judge tasks and in some cases made …
GitHub deprecated GPT-5.2 and GPT-5.2-Codex across most Copilot experiences on June 5, 2026, with GPT-5.5 and GPT-5.3-Codex as suggested replacements. The models remain available for Copilot code review, and Enterprise a…
Claude Code now includes a native PushNotification tool that allows the AI assistant to send alerts to a user's phone or Claude app when long-running tasks complete or require attention. The tool is designed for backgrou…
A team of developers built Thousand Token Wood, a multi-agent economic simulation where five AI-powered woodland creatures trade goods using a 3-billion-parameter Qwen2.5-3B model. The simulation, created for the Build S…
The Framework Desktop is a 4.5-liter mini tower built around AMD's Ryzen AI Max+ 395 "Strix Halo" chip with up to 128 GB of soldered unified memory, designed specifically for running large local LLMs like 70B-class model…
A developer has outlined the persistent technical limitations of large language models (LLMs), including hallucination, static knowledge cutoffs, statelessness, limited context windows, and compounding errors in multi-st…
Researchers have developed a Bayesian statistical framework for migrating production systems reliant on large language models when the underlying model reaches end-of-life. The approach calibrates automated evaluation me…
A new open-source workflow called "idea-to-build" guides founders and creators through a six-phase structured brainstorm that researches, critiques, and fact-checks raw ideas before handing them off to Claude Code as a b…
Meta, Llama 3 entegrasyonuyla Instagram ve WhatsApp'ı otonom çalışan dikey dağıtım kanallarına dönüştüren yeni bir yapay zeka asistanı sundu. Asistan, içerik üretenler ve küçük ölçekli girişimciler için platformun API ka…
Generative Engine Optimization (GEO) is emerging as a critical strategy for B2B companies in Latin America as executive-level buyers increasingly turn to AI platforms like ChatGPT and Gemini for vendor research instead o…
A developer's analysis of AI-generated code in the rsync utility found that Claude introduced subtle but significant bugs, raising concerns about AI reliability in critical codebases. Separately, researchers traced a pow…
Dropbox has unveiled Nova, an internal platform engineered to orchestrate and scale AI coding agents for tasks like code generation, bug fixing, and refactoring across its large codebase. OpenAI detailed the secure Windo…
GMKtec released the EVO-X2, a sub-$1,500 mini PC powered by AMD's Ryzen AI Max+ 395 processor with up to 128 GB of unified memory, enabling local operation of 70-billion-parameter AI models without a discrete GPU. The de…
Palantir CEO Alex Karp compared excessive AI token consumption to a "porn addiction" during a conference appearance, criticizing the "tokenmaxxing" trend as unproductive and masturbatory. Karp announced his company has d…
Researchers introduced two new consistency training methods, AttCT and MLPCT, which enforce output consistency in attention weights and MLP post-activations respectively, and applied all four existing and new methods to …
A March 2026 replication of the GSM-Symbolic study found that GPT-4o, Claude Opus 4.6, and Claude Haiku 4.5 no longer show catastrophic performance drops on confounded math problems when ambiguous samples are removed. Th…
The National Security Agency is using Anthropic's Mythos AI model for offensive cyber operations likely targeting China and Iran, with the company embedding engineers at the agency to customize the technology. The collab…
IBM Research launched Granite Libraries and Project Granite Switch, two tools that apply software engineering principles like modularity and plug-and-play components to large language models. The tools allow developers t…