OpenAI has plans to release an open-weight model in 2025

Hello everyone! Here’s what you need to know about AI today:

👉 OpenAI wants to seriously go open-source

👉 Apple adds Apple Intelligence to the Vision Pro headset

👉 Manus AI finally launched paid subscription plans

and many more!

📧 Did someone forward you this email? Subscribe here for free to get the latest AI news everyday!

Read time: 5 minutes

What’s going on: OpenAI has revealed plans to release its first open-weight language model since GPT-2, expected sometime in 2025, with a focus on integrating advanced reasoning capabilities. Unlike fully open-source models that share code and training data, this open-weight approach will make the model’s trained parameters public, allowing developers to analyze and adapt it for specific tasks without access to the original training materials.

What does it mean: OpenAI is on the edge of making a historical shift in its strategy. So far, they have kept their cutting-edge models like GPT-3 and GPT-4 behind closed APIs or subscription walls. Sam Altman claimed this release could happen as early as summer, following a thorough evaluation under OpenAI’s preparedness framework. The decision comes amid growing competition from rivals like DeepSeek and Meta’s Llama (with over 1 billion downloads), who have gained traction with open-source approaches.

More details: 

• OpenAI is engaging developers directly, starting with events in San Francisco and expanding to Europe and Asia-Pacific, to gather feedback and refine the model’s utility.

• Altman has acknowledged the company may have misstepped by not embracing open strategies sooner, though he notes it’s not a universal view within OpenAI nor their top priority.

• The new model, likened to the reasoning abilities of o3-mini, aims to balance accessibility with performance.

• If you’re interested in joining a feedback session with the OpenAI team, let them know by filling out this form.

• Want to see Sam Altman’s original post announcing this? Read his long post on X.

What’s going on: Apple has rolled out visionOS 2.4, introducing Apple Intelligence features to the Vision Pro headset. This release brings a suite of AI-driven tools to the mixed-reality device, including writing aids that allow users to rewrite, proofread, and summarize text across various applications. Additionally, the update incorporates Image Playground and Genmoji, enabling the creation of AI-generated images and custom emojis directly within the Vision Pro environment.

What does it mean: Finally Apple is taking AI more seriously! These new features aim to enhance productivity and creativity, leveraging natural language processing for tasks like searching photos by description or generating themed "Memory Movies" from a user’s media library.

More details: 

• The update also includes practical enhancements beyond Apple Intelligence, such as the new Apple Vision Pro app for iPhone, which lets users browse and transfer content like movies and TV shows to the headset.

• Other additions, like the Spatial Gallery app and improved Guest User functionality, expand content discovery and sharing options.

• While the AI tools mirror those already available on iOS, iPadOS, and macOS, their integration into visionOS 2.4 adapts them for spatial computing, though no Vision Pro-exclusive AI features were introduced in this update.

• Want to learn more about the details? Read Apple’s official announcement here.

🚀 Manus AI, the viral chinese AI agent platform, has launched paid subscription plans with added features, an iOS app, and upgraded its AI model to Claude 3.7 Sonnet.

🧠 Neurologyca, a Spanish company, has developed an AI system that analyzes facial expressions to detect subtle emotional patterns and micro-movements, potentially enabling early detection of neurological issues like strokes and brain hemorrhages, with applications also extending to education, security, and defense.

📄 A new paper from the AI Disclosures Project alleges that OpenAI trained its GPT-4o model on paywalled books from O'Reilly Media without a licensing agreement, as GPT-4o demonstrated a stronger recognition of paywalled O'Reilly book content compared to earlier models.

🚶‍♀️ Meta's VP of AI Research, Joelle Pineau, announced her departure from the company in May after more than two years overseeing FAIR (Facebook Artificial Intelligence Research) division.

🐌 Sam Altman, stated that the popularity of their new image-generation tool in ChatGPT will cause product delays, service disruptions, and slower performance as the company addresses capacity challenges, with ChatGPT now having 500 million weekly users and 20 million paying subscribers.

💰 Qualcomm has acquired the generative AI division of VinAI, a Vietnamese AI research company, for an undisclosed amount to expand its AI tooling sector and strengthen its ability to deliver cutting-edge AI solutions.

AI + Investment advice

LinkedIn - Product Manager, Network Growth AI

Coactive - MLOps Engineer

Aurora - ML Software Engineer, Behavior Planning

ManTech - Senior Data Scientist

Thank you for staying with us like always! If you are not subscribed, subscribe here for free to get more of these emails in your inbox! Cheers!