💜 AI News

The latest AI news, processed and simplified so you can stay up-to-date.

Hugging Face launches MCP to accelerate research

Hugging Face introduces the Model Context Protocol (MCP) so AI agents can orchestrate research tools. The article explains three levels of discovery, showcases the `research-tracker-mcp` demo, gives quick setup steps and outlines risks and best practices for reliable use....

Anthropic updates its AI Usage Policy

Anthropic updates its `Usage Policy` on August 15, 2025, with changes that take effect on September 15, 2025. The updates clarify rules on cybersecuri...

Meta and WRI use DINOv3 to count trees from space

Meta and the World Resources Institute adapted `DINOv3` to count trees from satellite images and published open maps and models. The technology speeds...

Meta DINO powers JPL and NASA explorer robots

Meta launches `DINOv3`, a self-supervised visual backbone that JPL is using to improve vision and autonomy in explorer robots. It helps with tasks lik...

Meta introduces DINOv3: label-free AI vision

Meta announces DINOv3, a new iteration of self-supervised vision models. The original blog required login, so this piece synthesizes public informatio...

Gemma 3 270M: compact AI model for devices

DeepMind presents Gemma 3 270M, a 270M-parameter model focused on energy efficiency and quick fine-tuning for specific tasks. It's ideal for on-device...

TextQuests puts LLMs to the test with text games

TextQuests uses 25 classic Infocom text games to benchmark long-term reasoning and memory in LLMs. Results reveal failures in very long contexts, hall...

OpenAI asks California to harmonize AI regulation

OpenAI sent a letter to Gov. Gavin Newsom asking California to harmonize its AI regulation with federal and global standards, avoiding duplicated stat...

Basis transforms accounting with GPT‑5 and AI agents

Basis uses `GPT‑5` and a multi‑agent architecture to automate accounting tasks, offering up to 30% time savings and reviewable explanations. Benchmark...

Anthropic expands Claude Sonnet 4 context to 1M tokens

Anthropic announces Claude Sonnet 4 supports up to 1M tokens of context (public beta), enabling large codebase and document analysis in one request. A...

Anthropic details safeguards for Claude

Anthropic publishes how its Safeguards team designs policy, training, testing and real-time detection to minimize abuse of Claude. The approach combin...

Anthropic offers Claude to the entire U.S. government for $1

Anthropic offers Claude (Enterprise and Government) to all three branches of the U.S. government for $1, with technical support and a FedRAMP High–cer...

OpenAI introduces GPT-5 for code and design

OpenAI released GPT‑5 (August 7, 2025), a model optimized for code, design and agentic tasks: improvements on programming benchmarks, preferred fronte...

GPT-5 drives medical research with advanced AI

OpenAI presents `GPT-5` with specific improvements for medical research and healthcare use: better clinical reasoning, fewer hallucinations, and highe...

GPT-5 for creative writing: OpenAI releases new model

OpenAI published on August 7, 2025 an article about GPT‑5 focused on creative writing. The model promises help with plots, characters, and styles; the...

OpenAI introduces GPT-5: Cursor, AI with contextual control

OpenAI announced GPT-5: Cursor on August 7, 2025 — a version designed to work directly in the context where you are (documents, code, tabs). This piec...

GPT‑5 from OpenAI: unified AI with 'thinking' built in

OpenAI unveiled GPT‑5 on August 7, 2025: a unified model that decides when to answer quickly and when to 'think' more to deliver expert-level response...

GPT-5 arrives at Amgen: AI that speeds scientific decisions

OpenAI launched GPT-5 (August 7, 2025) and Amgen was among the first to evaluate it; the company reports improvements in accuracy and output quality f...

Perch: AI that accelerates bioacoustics to save species

DeepMind released an updated open-source version of Perch (August 7, 2025), an AI model that analyzes natural sound recordings to identify species and...

OpenAI unveils GPT-5 and redefines work

OpenAI announced GPT-5 on August 7, 2025, with improvements in reasoning, speed and context handling. Initial rollout began for Team users, the API is...

OpenAI launches GPT‑5 for developers: code and agents

OpenAI launches GPT‑5 for developers focused on code and agents. It adds reasoning and verbosity controls, new custom tools, and strong benchmark gain...

Llama powers Biofy against antibiotic resistance

Biofy uses `Nanopore` sequencing and the `Abby Recommender` platform to identify bacteria and their resistances in under 4 hours (previously ~5 days),...

OpenAI introduces GPT-5: unified model with reasoning

OpenAI launches GPT‑5 as a unified model with automatic reasoning and improvements in writing, code, health and factuality. It becomes the default in ...

Claude Code adds automated security reviews

Anthropic announced on August 6, 2025 that Claude Code now offers `/security-review` from the terminal and a GitHub Action to automatically analyze pu...

Anthropic publishes a framework for safe AI agents

Anthropic published on August 4, 2025 an initial framework for the responsible development of AI agents, focused on human control, transparency, value...

Anthropic launches Claude Opus 4.1: improvements in code and agents

Anthropic launched Claude Opus 4.1 on August 5, 2025 — an update focused on agents, reasoning and real-world programming that improves code performanc...

Anthropic launches program on AI's economic impact

On June 27, 2025, Anthropic announced the Economic Futures Program to fund research, symposia and economic measurement studying AI's impact on work an...

DeepMind introduces Genie 3, real-time interactive AI worlds

DeepMind introduced `Genie 3` (Aug 5, 2025), a world model that generates interactive worlds in real time at ~24 fps and ~720p, maintaining coherence ...

OpenAI launches open weights models to democratize AI

On August 5, 2025, OpenAI released its most capable models as open weights, integrated them with OpenAI for Countries, and extended support to nonprof...

OpenAI releases gpt-oss: open models and safety

OpenAI released two open models, `gpt-oss-120b` and `gpt-oss-20b`, under Apache 2.0, designed for reasoning and customization. OpenAI’s safety evaluat...

OpenAI introduces GPT‑OSS: open models 120B and 20B

OpenAI releases GPT‑OSS: two reasoning models with open weights (120b and 20b) under Apache 2.0, designed to run and be customized on your own infrast...

DeepMind and Kaggle launch Game Arena to measure AI intelligence

DeepMind and Kaggle launch Game Arena, a public platform that evaluates AI models by pitting them in strategic games with open rules and an `all-play-...

Deep Think arrives in Gemini: new AI to solve problems

Google/DeepMind released Deep Think —a Gemini 2.5 variant that uses parallel thinking— in the Gemini app for Google AI Ultra subscribers (announcement...

Upwork integrates Meta’s Llama to help freelancers

Upwork integrated Meta’s Llama into its assistant `Uma` to improve proposals, interviews, and meeting summaries (announced July 23, 2025). The update ...

AWS and Meta launch program for startups using Llama

AWS and Meta launched a six-month program for 30 startups using Llama, offering up to $200,000 in AWS credits and joint technical support. Targeted at...

Meta introduces Seamless for more human conversations

Meta introduced Seamless Interaction, a project that pairs a `dataset` of over 4,000 hours with audiovisual models to generate gestures, expressions a...

Llama 2 helps ANZ Bank improve operational efficiency

Meta publishes that `Llama` helps ANZ Bank improve efficiency, but the blog post requires login so I couldn't read it fully. ANZ publicly uses AI tool...

Stay up to date!

Receive practical guides, fact-checks and AI analysis straight to your inbox, no technical jargon or fluff.