// a daily digest from the AI world
New models, tools and announcements every day, always with a source link. This digest is assembled by an AI agent following my rules, and it's the same briefing I use myself to keep up with AI without burning time.
French AI lab Mistral AI has released Mistral Large 4, a large multimodal model with a trillion parameters. It aims to leapfrog both American and Chinese rivals. For the European AI scene it is an important signal.
The new EmbeddingGemma 2 embedding model ships under the open Apache 2.0 license. For embeddings this makes a lot of sense, since applications usually compute thousands or millions of vectors and store them for later comparison. An open model is not tied to a single provider.
OpenAI is adding an invisible, machine-readable watermark to text from ChatGPT and Codex, at first only for users in the European Union to comply with the AI Act. The company admits that editing the text can make the marks harder to detect.
Anthropic is launching a program for startups with a free year of Claude Team and $1,000 in token credits. The company says the benefits of AI will reach most people through the companies that build on top of models, not through the models alone.
Starting October 9, Gemini users on the free plan will be limited to the Flash Lite model. Until now they could also choose Flash and Pro. The standard Flash model will require a Google AI Plus subscription at $4.99 per month.
OpenAI has published new results on open problems in mathematics from an internal frontier model. It is sharing Lean proof formalizations and research details on GitHub. That lets the community verify the work.
This section writes, assembles and publishes itself every morning, from source selection to deployment. If you'd like a quiet helper like this in your company, let's start with an initial consultation. The fee is credited toward the project if we work together.