Hello, Starters!

The AI landscape has more sudden turns than a rollercoaster. It's easy to be caught off guard by them. Luckily, we're here to clear up any doubts you may have about these unexpected changes!

Here’s what you’ll find today:

• Apple and Microsoft are off OpenAI's board

• Huffington and Altman announce Thrive AI Health

• DeepMind’s JEST method for faster training

• SenseTime unveils a new multimodal model

• Meta launches MobileLLM

• And more.

👋 Apple and Microsoft are off OpenAI's board (1 min)

There has been another sudden change on OpenAI's board, and this time it involves two key players, Microsoft and Apple. They're no longer taking part as observers. After several reports, this move has been confirmed by OpenAI, claiming that they're shifting into a new approach with both companies, which includes regular stakeholder meetings.

Microsoft joined OpenAI's board almost eight months ago, after the shakeup the company faced during Altman's departure. Apple, on the other hand, was just recently in talks to grab a seat as an observer, but now, the companies are changing the course of their collaboration. Allegedly, this is a response to rising antitrust concerns.

🩺 Huffington and Altman announce Thrive AI Health (7 min)

There are many ways in which AI can help to improve health care; it is a topic that we've discussed a few times already. Sam Altman is aware of this, and that's why OpenAI has been involved in several initiatives that follow this notion. The latest is a joint effort with Thrive Global and its CEO, Arianna Huffington, which aims to launch a customised AI health coach through a mobile app, focused on changes to daily behaviours aimed at improving our health.

Thrive AI Health will be the result of Altman and Huffington's vision, with a methodology that introduces the concept of Microsteps, which mixes our preferences and routines with AI to come up with specific actions we can take to have a healthier life.

💥 DeepMind’s JEST method for faster training (1 min)

A research team at Google DeepMind has come up with JEST (Joint Example Selection Technique), a method that reduces the training time for multimodal AI models. They've achieved this feat by using two AI models to refine the subsets of data used for training based on their learnability.

Usually, how multimodal models link images and text consists of correspondence, where the pairs with the least relevance are left behind to maximise the correspondence of related ones. However, each batch of data might differ in quality. This is where JEST's algorithm comes in handy since it selects the most useful data.

🤖There's a new multimodal AI model that's suited to defying GPT-4o's capabilities. We're talking about SenseTime's SenseNova. Its latest 5.5 version has been claimed as China's first real-time multimodal model. Users can interact with it just by talking, and SenseNova easily processes audio, image, text, and video data.

📲Keeping up with the trend of more compact LLMs, researchers at Meta AI have developed MobileLLM, an optimised language model that's able to function without requiring much computational resources, which makes it perfect for smartphones and similar devices.

⚡️Quick links

MInference 1.0: Accelerating Pre-filling for Long-Context LLMs via Dynamic Sparse Attention

RouteLLM: A framework for serving and evaluating LLM routers

Artifacts made with Claude can now be published and shared

Ex-Microsoft engineer’s AI video startup scores $60M from top VCs, Jared Leto

The Washington Post debuts AI chatbot

Thank you for reading!