Good morning. It’s Monday, October 7th.

Did you know: On this day in 2003, Nokia released the N-Gage: a mobile phone that doubled as a handheld game console. It is considered a failure by the tech press, and was discontinued a few years later.

In today’s email:

• Meta’s Movie Gen

• Apple Intelligence Update

• Rationalyst Paper

• 5 New AI Tools

• Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

Meta enters AI video wars with powerful Movie Gen set to hit Instagram in 2025

Meta has announced the upcoming launch of Movie Gen, a suite of generative multimodal AI models designed for video and audio creation, set to debut on Instagram in 2025.

This advanced system features four key models: Movie Gen Video, a 30 billion-parameter text-to-video generator; Movie Gen Audio, a 13 billion-parameter audio generator; Personalized Movie Gen Video, which creates custom videos using users’ images; and Movie Gen Edit, focusing on precise video modifications.

Users can produce HD videos lasting up to 16 seconds with 48kHz audio through simple text prompts. Meta's approach combines traditional diffusion model techniques with a new method called Flow Matching, enhancing both efficiency and output quality. The training utilized vast datasets, although concerns about data sourcing have been raised.

Meta intends to democratize access to these advanced tools for both casual users and professional creators, with ongoing improvements planned for user experience and technological capabilities. Read more.

Apple releases Depth Pro, an AI model that rewrites the rules of 3D vision

Your iPhone 16 may get its first Apple Intelligence features later this month

Apple is gearing up to introduce AI features to the iPhone 16 with the upcoming iOS 18.1 update, slated for release on October 28, according to Bloomberg’s Mark Gurman. This update promises to roll out AI-powered notification summaries, offering a glimpse into the future of streamlined user engagement.

Among other nifty enhancements in the beta, Siri will now boast improved comprehension, effortlessly deciphering even the most garbled queries. Users can also expect a vibrant rainbow-colored edge graphic when invoking Siri, adding a splash of flair to voice commands.

While Apple Intelligence is limited to the iPhone 15 Pro and later models, it’s already available in beta on Macs and iPads with Apple silicon (M1 and above). However, some marquee features, like Siri's multi-app task execution, are still on the horizon, possibly arriving in March 2025, with fun extras like Apple’s Image Playground expected by December. Read more.

RATIONALYST: How implicit rationales improve AI reasoning

Researchers at Johns Hopkins University have introduced RATIONALYST, an AI model designed to enhance the reasoning capabilities of large language models by utilizing implicit rationales derived from unlabeled text data. This innovative system generates and filters justifications based on example prompts, capturing essential reasoning patterns for new texts.

Trained on approximately 79,000 implicit justifications, RATIONALYST monitors step-by-step problem solutions, generating reasoning for each stage to guide the selection of optimal next steps. In evaluations across various reasoning tasks, RATIONALYST achieved an average accuracy improvement of 3.9 percent, outperforming larger verifier models, including GPT-4.

The researchers view this method as a viable strategy for boosting both the interpretability and performance of language models in complex reasoning scenarios, with plans for future enhancements involving stronger models and more extensive datasets. The model's code is publicly available on GitHub for further exploration. Read more.

• Qualcomm debuts AI-optimized A7 Elite chip for powering Wi-Fi systems

• Big Tech has cozied up to nuclear energy

• Anthropic boosts RAG accuracy with context-aware retrieval

• Gartner Says Generative AI will Require 80% of Engineering Workforce to Upskill Through 2027

• Renowned mathematician Terence Tao envisions AI-powered "industrial-scale mathematics"

• South Korea set to criminalize possessing or watching sexually explicit deepfake videos

📄 Loong: Generating Minute-level Long Videos with Autoregressive Language Models

📄 LLaVA-Critic: Learning to Evaluate Multimodal Models ​​

📄 Not All LLM Reasoners Are Created Equal  

📄 Addition is All You Need for Energy-efficient Language Models

📄 Revisit Large-Scale Image-Caption Data in Pre-training Multimodal Foundation Models

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email!