Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Apple confirmed it does not use Google's Gemini models, Google Search, or Google Assistant infrastructure to power Siri or Apple Intelligence, despite a partnership involving Nvidia GPUs hosted in Google's cloud. Senior …
NVIDIA and MiniMax released the M3 multimodal AI model on NVIDIA accelerated infrastructure, including Blackwell GPUs, enabling long-context reasoning and agentic workflows. The 428-billion parameter Mixture-of-Experts m…
A developer created a script called 'mypedia' that uses ChatGPT to generate short, balanced reference pages in the style of Wikipedia articles. The script takes a name as input, prompts ChatGPT to produce a neutral summa…
Researchers have introduced GraphRAG, a graph-based approach to retrieval-augmented generation that enables large language models to answer global, query-focused summarization questions over entire text corpora. The meth…
Developer Yuga Hashimoto released LocalAnt, a local-first MCP gateway that gives ChatGPT structured access to a user's local machine without exposing a raw shell. The tool acts as a permissioned middle layer, allowing Ch…
Kilo AI launched REVIEWS.md and Code Review Memory for its automated code review tool, allowing teams to define repository-specific standards in a Markdown file and enabling the AI agent to learn from past feedback to ad…
Anthropic CEO Dario Amodei on February 26 refused Pentagon demands to remove safety guardrails from the Claude AI model for military use, citing risks of mass domestic surveillance and fully autonomous lethal weapons. Th…
A data scientist achieved a 40-65% cost reduction by integrating DeepSeek models into a Next.js application, replacing an OpenAI-only stack across 12,000 production requests over six weeks. The project, which served a do…
A developer wasted three days building a complex notification system with WebSocket updates and badge counts, only to learn the product team simply wanted an email when a task was assigned. To prevent this costly mistake…
Claude uses web search in only 36.6% of prompts compared to ChatGPT's 90%, according to data from a Zero Click by Profound session shared by Jonathan Clark. The AI model relies directly on Brave Search's top 10 results w…
Grok 4.3 defeated GPT 5.4 by a score of 36.0 to 34.0 in a narrow contest focused on format compliance. GPT 5.4 outperformed on technical tasks like Python redact logs and IPv4 handling, but Grok 4.3 secured victory by ma…
Ollama, LM Studio, and llama.cpp now support native Anthropic Messages API connections, enabling Claude Code to run entirely on local models for code completion, refactoring, and debugging tasks. By setting environment v…
A new study published in *PNAS* found that large language models can be tricked into bypassing their safety guardrails using classic human psychological persuasion techniques, such as appeals to authority, scarcity, and …
Mistral is in early-stage funding talks to raise approximately €3bn ($3.5bn) at a valuation of roughly €20bn, nearly doubling its September valuation of €11.7bn. The Paris-based AI startup, positioning itself as Europe's…
French AI startup Mistral AI is negotiating a new funding round of approximately 3 billion euros, which would value the company at around 20 billion euros. The company, positioning itself as a European alternative to US …
A new open-source tool called cost-xray captures the actual local API traffic for AI coding agents Claude Code and Codex, attributing tokens and costs down to individual sources like system prompts, tool schemas, and MCP…
A developer using Claude 5 Fable (Ultracode) generated a fully explorable 3D starship demo in a single prompt using Three.js, complete with a working cockpit, crew quarters, dynamic lighting, and interactive sleep/eat me…
Meta launched Muse Spark on April 8, a new AI model with selectable operational modes for different cognitive demands, marking its first major release from the Muse family and a strategic shift from open-source to hosted…
Anthropic's Fable 5 model has become the company's most "honest" model to date, frequently telling users it is being honest in its responses. A user who indexed 202,381 messages from local Claude Code sessions over the p…
RedAnalytica has released RedNotebook AI, an open-source AI data notebook that integrates with Trino, DuckDB, and 11 other SQL engines to provide a unified workspace for querying, visualizing, and exploring data. The too…