🌊 Flood of Mini AI Models

PLUS: Meta Halts EU AI Releases

Welcome back!

The latest AI battleground is small models. After OpenAI, Mistral, Hugging Face, and Apple have just unveiled their new "mini" AI models, for efficient, privacy-focused, and cost-effective local applications. Let’s unpack…

Today’s Summary:

  • 🤖 Mistral, Hugging Face, Apple launch compact models

  • 🛑 Meta halts EU release of advanced AI

  • đź’¬ OpenAI discusses custom chips with Broadcom

  • 📱 Apple, Salesforce respond to YouTube training claims

  • 🛡️ OpenAI, Google, Microsoft join Coalition for Secure AI

  • 🌟 Elon Musk promises world's most powerful AI by December

  • 2 new tools

TOP STORY

Mistral, Hugging Face, and Apple launch their “Mini” AI models

The Summary: In response to OpenAI “mini” model release, Mistral AI with NVIDIA, Hugging Face, and Apple have unveiled their own open-source compact AI models. Mistral-NeMo, Hugging Face SmolLM, and Apple DCLM models showcase impressive capabilities despite their reduced size. These developments bring advanced AI closer to edge devices, enhancing privacy and reducing computational costs.

Key details:

  • Mistral-Nemo: 12B parameters, 128k token context window, collaboration with NVIDIA, Apache license

  • Hugging Face SmolLM: 135M, 360M, and 1.7B parameter versions suitable for smartphones, Apache license

  • Apple DCLM-7B: outperforms Mistral-7B, while DCLM-1.4B outperforms SmolLM-1.7B. Dataset also released. This is not “Apple Intelligence” model.

Why it matters: The release of these open-source, compact AI models shows a rising trend towards developing AI running on local hardware. This addresses growing concerns about cloud environmental impact, data privacy, and the need for faster and cheaper processing. As these small models become more powerful, we may see a surge in AI applications running locally.

META

Meta halts EU release of advanced AI

The Summary: Meta has decided not to release its upcoming multimodal AI model in the European Union. The company cites "unpredictable" privacy regulations as the reason for this move. This decision affects Meta's plans to integrate AI into products like smartphones and Ray-Ban smart glasses. The company's stance highlights growing tensions between big tech firms and EU regulators over data protection and AI governance.

Key details:

  • Meta's multimodal Llama model can process text, video, images, and audio

  • Decision based on concerns about complying with regulations

  • Text-based versions of Llama will still be available in the EU

  • Meta also suspended AI tool use in Brazil due to privacy concerns

Why it matters: This move may create a technology gap between EU and non-EU consumers. It could pressure EU regulators to clarify AI regulations, potentially leading to a compromise between innovation and data protection. The decision may also influence other tech companies' strategies in the EU market.

OPENAI

OpenAI in talks with Broadcom for custom AI chips

The Summary: OpenAI is discussing the development of a new AI server chip with semiconductor designers, including Broadcom. This move aims to reduce dependence on Nvidia and strengthen OpenAI's supply chain. The talks with Broadcom, while in early stages, reflect OpenAI's strategy to maintain its edge in the competitive AI landscape.

Image: Broadcom

Key details:

  • OpenAI chip team includes former Google TPU developers

  • Discussions involve multiple industry and government stakeholders

  • New chip development could take until at least 2026

  • OpenAI seeks to increase access to AI infrastructure

Why it matters: This initiative highlights the growing importance of specialized hardware in AI development. As AI companies compete for limited chip supplies, custom chip solutions could reshape the industry and reduce dependency on Nvidia.

QUICK NEWS

Quick news

TOOLS

🥇 New tools

  • Flot - Your personal knowledge base to remember everything

  • Supermemory - AI second brain for all your saved stuff

That’s all for today!

If you liked the newsletter, share it with your friends and colleagues by sending them this link: https://thesummary.ai/