Apple and Meta Unveil New AI Innovations

ALSO: Exploring AI's memory challenges and Nvidia's leap into physical AI

Author

The Chirp AI
July 30, 2024

**

Welcome back, AI Aficionados

Apple's latest iPhone AI preview is here, promising to revolutionize user interaction, while Meta's new video vision AI, presented by Zuckerberg and Nvidia's Jensen Huang, is setting new standards in the industry.

 Today’s Insights:

  • Apple releases first preview of its long-awaited iPhone AI

  • Challenges of making AI models forget undesirable data

  • Meta’s latest video vision AI unveiled

  • 4 AI Tools to Turbocharge Your Productivity

  • AI Case Studies

Apple unveils Apple Intelligence in iOS 18.1 Developer beta

Source: Apple

Today Apple released the first version of Apple Intelligence.
The new software called Apple Intelligence was released in the developer beta of iOS 18.1. For now, it is only available to registered Apple developers who pay $99 per year.

Key Points:

  • Apple released the first version of Apple Intelligence on Monday.

  • The new software, called Apple Intelligence, is available in the developer beta of iOS 18.1.

  • Currently, it is only accessible to registered Apple developers.

Details: 

Apple Intelligence, introduced during Apple's WWDC 2024 in Cupertino, California, on June 10, 2024, is a suite of artificial intelligence features designed to improve Siri, automatically generate emails and images, and sort notifications. The new software is available in the developer beta of iOS 18.1 and similar releases for iPad and Mac. It is currently only available to registered Apple developers, who must pay $99 per year.

What Users Can Expect:

  • New Siri look: Makes the edge of the phone glow.

  • Improved Siri commands: Understands commands even when the speaker stumbles.

  • Troubleshooting: Siri can answer troubleshooting questions about Apple products.

  • Photo and video enhancements: Better photo search and movie creation.

  • AI-generated summaries: For Mail, Messages, and voicemail transcriptions.

  • Writing Tools: Apple’s next-generation service.

What's Missing:

  • Image and emoji generation: These features will be rolled out over the next year.

  • Automated photo clean-up: Not included yet.

  • Further, Siri improvements include the ability to use personal information and take action inside apps.

  • OpenAI ChatGPT integration: Will be added later.

Why It Matters: Investors hope the tight integration of AI with Apple’s operating system can spur a big wave of upgrades in the coming years, especially since the system will only work on the iPhone 15 Pro and iPhone 15 Pro Max and newer. Bank of America analyst Wamsi Mohan expects this iPhone cycle to remain strong for longer as AI feature sets improve in the 2025 iPhone.

Making AI models ‘forget’ undesirable data hurts their performance

Source: Techcrunch

Unlearning techniques in AI models can degrade their performance.
So-called “unlearning” techniques make a generative AI model forget specific and undesirable info it picked up from training data, like sensitive private data or copyrighted material.

Key Points:

  • Unlearning techniques are designed to make AI models forget undesirable data.

  • These techniques can significantly degrade the performance of AI models.

Details: A new study co-authored by researchers at the University of Washington (UW), Princeton, the University of Chicago, USC, and Google found that current unlearning techniques tend to degrade AI models, often to the point where they’re unusable. These findings highlight the double-edged nature of unlearning methods.

What the Study Found:

  • Popular unlearning techniques degrade models: The study found that while these techniques can make AI models forget certain information, they also hurt the models' general question-answering capabilities.

  • Benchmark testing: Researchers devised a benchmark called MUSE (Machine Unlearning Six-way Evaluation) to test the effectiveness of unlearning algorithms. Scoring well on MUSE requires making a model forget specific data without significant loss of utility.

Challenges in Unlearning:

  • Interconnected knowledge: Knowledge in AI models is intricately entangled. For example, removing information about Harry Potter books also affects the model's knowledge of freely available content from the Harry Potter Wiki.

  • No efficient methods yet: Current unlearning methods are not yet efficient or practical for real-world deployment.

Why It Matters: The ability to make AI models forget specific data is crucial for addressing privacy concerns and complying with legal requirements. However, the degradation in model performance poses a significant challenge, highlighting the need for further research and development in this area.

Zuckerberg touts Meta’s latest video vision AI with Nvidia CEO Jensen Huang

Meta's Segment Anything 2 moves into video.
Meta had a palpable hit last year with Segment Anything, a machine-learning model that could quickly and reliably identify and outline just about anything in an image. The sequel, Segment Anything 2 (SA2), takes the model to the video domain.

Key Points:

  • Segment Anything 2 (SA2) applies natively to video.

  • CEO Mark Zuckerberg debuted SA2 at SIGGRAPH 2024 with Nvidia CEO Jensen Huang.

Details: Segmentation allows a vision model to identify parts of an image or video, like distinguishing a dog from a tree. SA2 enhances this capability for video, making it more efficient than running the first model on individual frames. This advancement is crucial for scientific studies and other applications that require detailed video analysis.

Why It Matters: Processing video is computationally demanding. The ability of SA2 to handle this without overwhelming data centers showcases significant advances in AI efficiency. Meta has released a large, annotated database of 50,000 videos for training purposes. This open model and tools like PyTorch and LLaMa reinforce Meta's commitment to advancing accessible AI technologies.

Tech Quick Bites

4 AI Tools to Turbocharge Your Productivity

🤖 Fillout
Fillout is a powerful AI-driven form builder designed to streamline data collection and analysis. It offers customizable templates and real-time data processing, making it ideal for businesses and organizations looking to enhance their data management strategies.

🤖 Clockwise
Clockwise is an AI-powered time management tool that optimizes your schedule by automatically arranging meetings, focus time, and breaks. It's designed to help professionals maximize productivity and maintain a balanced work-life schedule.

🤖 Decktopus
Decktopus leverages AI to create engaging and professional presentations effortlessly. With a variety of templates and smart design suggestions, it helps users create visually appealing slideshows in minutes.

🤖 Plus AI
Plus AI integrates with existing software to provide advanced data analytics and insights. It's designed to help businesses make informed decisions by analyzing trends and predicting future outcomes.

AI Case Studies

  • Accenture’s Strategic Embrace of AI: Accenture has integrated AI across its operations through initiatives like Accenture LearnVantage and collaborations with AI startups. These efforts focus on upskilling the workforce and delivering innovative AI solutions to clients. Read more

  • Amazon’s Evolution as an AI-Driven Powerhouse: Amazon uses AI extensively for personalized product recommendations, supply chain optimization, fraud detection, and customer service through Alexa. AI plays a crucial role in approximately 35% of Amazon’s total sales. Read more​ (AIX | AI Expert Network)

Thanks for reading today’s edition.

Stay curious and keep exploring the ever-evolving world of AI. Until next time!

THE CHIRP AI team