Hello đź‘‹ Apple Intelligence (AI)

Along with: Childish Elon Musk games!

Hey there, 

As a creator, what I appreciate most about Apple is its commitment to starting with the user experience. Their design and presentation revolve around enhancing this experience, using technology as a means to deliver it seamlessly.

In contrast, Google takes an almost diametrically opposite view to mention “AI” even when it feels like a simpler decision engine!

In that context - it was refreshing to see Apple announce its Intelligence - the child in me wants to just play around with the Play notes and visualize the parabolic and hyperbolic relationships through them! Sadly - I would need to buy an M series iPad for it.

Apart from Apple Intelligence, there are interesting updates on Claude, Kling, Microsoft, and OpenAI! Let’s run through them.

So coming to the newsletter!!

What would be the format? Every week, we will break the newsletter into the following sections:

  • The Input - All about recent developments in AI

  • The Tools - Interesting finds and launches

  • The Algorithm - Resources for learning

  • The Output - Our reflection 

Table of Contents

WWDC 24 finally took place, and the highlight was undoubtedly AI - Apple Intelligence.

Unlike Microsoft, Google, and OpenAI, Apple isn’t emphasizing the number of parameters in its model. Instead, it introduced its own foundation model focused on on-device processing. Specialized tasks are handled by its large foundation model through Private Cloud Compute.

For tasks that are too complex, Apple explicitly seeks user permission and utilizes ChatGPT for assistance.

Apple has made significant strides in privacy. All data is encrypted, ensuring no unauthorized access.

Apple Intelligence is a personal intelligence system that brings generative AI capabilities to the iPhone, iPad, and Mac. It understands user behavior and provides personalized intelligence by combining AI models with personal context.

Capabilities of Apple Intelligence

  • Personalized Assistance: Provides tailored responses to user queries by considering the user's context and preferences.

  • Enhanced Siri: Upgraded with generative AI, enabling it to perform complex tasks such as summarizing news articles, editing photos, and generating email responses.

  • Generative AI Tools: Allows users to create text and images based on prompts, including custom emojis through the Genmoji tool.

  • On-Device Processing: Most AI tasks are handled on-device, ensuring privacy and security. For more demanding tasks, Apple uses private cloud computing to maintain data confidentiality.

  • Integration with Apps: Summarizes notifications, categorizes emails, provides intelligent auto-replies in Mail and Messages, enhances photo editing in the Photos app, and transcribes Voice Memos.

  • Developer Tools: Xcode features AI-powered code completion and other generative AI tools to assist developers. (source)

Usually, companies create AI models and aim to train them for harm avoidance. But being truly helpful or likable is about more than just avoiding harm. It also involves being curious, truthful, understanding, and even funny- qualities we appreciate in people.

Recognizing this - the developers introduced "character training" in Claude 3. This approach is more than just an alignment technique or a feature to enhance user experience; it fundamentally shapes how the model interacts with complex and varied human values and situations. 

This training helps the AI understand different perspectives and react in a more sophisticated and human-like way to various situations. (source)

Elon Musk has withdrawn his lawsuit against OpenAI co-founders Sam Altman and Greg Brockman just a day after criticizing their partnership with Apple.

The lawsuit, filed earlier this year, accused OpenAI of breaching an agreement and shifting from its initial mission of developing AGI for humanity's benefit to becoming a for-profit entity dominated by Microsoft. Legal experts had doubted the case's foundation due to the lack of a formal agreement. (source)

At Apple's recent Worldwide Developers Conference (WWDC), the company announced a significant partnership with OpenAI to integrate ChatGPT into Apple devices, enhancing the capabilities of iOS, iPadOS, and macOS with GPT-4o-powered ChatGPT.

This move drew sharp criticism from Elon Musk, who expressed concerns about privacy and security implications.

He even tweeted threats to ban Apple devices at his companies and demanded that visitor devices be stored in Faraday cages to block all communication. Following this, Musk hinted at a potential partnership with Samsung to develop an "X phone," further complicating the tech landscape. (source)

Several people clarified that the permission will be explicitly asked and not have a native integration..

Kling AI, developed by China’s Kuaishou Technology, has emerged as a leader in the AI-driven video creation sector, surpassing competitors like OpenAI's Sora model.

This text-to-video generation model is capable of creating highly realistic videos up to two minutes long, using advanced 3D reconstruction technology to achieve lifelike visuals and detailed, dynamic scenes.

Kling AI's capabilities include generating 1080p resolution videos at 30 frames per second, and its use of a 3D Variational Autoencoder enhances the realism of facial and body movements.

It offers more flexibility than its rivals by producing longer videos and has an open-access model with regional restrictions, providing a significant competitive advantage in the global AI market. (source)

Stability AI has launched a new open AI model called Stable Audio Open.

This model, which can generate sound and music snippets based on text descriptions, was trained on around 486,000 samples from royalty-free music libraries.

While Stable Audio Open enables the generation of audio elements like drum beats and ambient sounds, it's restricted to non-commercial, short clips without full songs or vocals, and faces biases and limitations in diversity and language. (source)

Hugging Face introduced its new robot, "Le Robot," by posting a video on X that showcased Reachy2, the first humanoid robot from this initiative, developed in collaboration with Pollen Robotics.

Reachy2, which can perform household chores safely around humans and pets, was initially trained through teleoperation by a human using VR and then learned to operate independently using machine learning algorithms analyzing short video clips of these sessions.

Hugging Face has made the training dataset and model openly available, promoting open-source development in robotics. (source)

Microsoft is retiring its GPT Builder tool and support for Copilot GPTs from the Copilot Pro subscription between July 10 and July 14, 2024.

This change will remove users' ability to create and access their GPTs. Introduced three months ago, GPT Builder allowed users to create mini chatbots for specific tasks.

Microsoft has not provided reasons for this decision but will focus on core Copilot functionalities. Users are advised to save their GPT instructions before the feature is discontinued, as all associated data will be deleted. (source)

At a business event, Meta announced its latest enhancement for WhatsApp business, focusing on integrating AI tools to streamline customer interactions and promotional activities.

AI is being trained to autonomously handle common customer inquiries on WhatsApp, facilitating quicker and more efficient customer service.

Additionally, this AI integration extends to aiding businesses in creating targeted ads on Facebook and Instagram, managing cart abandonment, and offering timely discounts. (source)

Researchers have developed new methods to enhance the interpretability of neural networks, particularly focusing on language models like GPT-4.

By using sparse autoencoders, they have identified 16 million features that can potentially help in understanding the complex neural activities within these models.

These features, which show patterns of activity related to specific human-understandable concepts, are intended to make it easier for AI to assist in answering queries and performing tasks by mimicking human reasoning processes more closely. (source)

OpenAI has strengthened its leadership by appointing Sarah Friar as CFO and Kevin Weil as CPO, aiming to enhance its AI product revenue and address safety concerns.

Friar, former CEO of Nextdoor, and Weil, an ex-executive at Instagram, join during a pivotal time as OpenAI navigates commercial expansion and competitive challenges while reaffirming its commitment to safe AI development. (source)

Tool: iplan.ai

Planning and organizing events can be a complex and time-consuming task.  iplan.ai offers an intelligent platform to streamline event planning, making it efficient and stress-free.

Problem Statement: Suppose you're an event planner managing multiple events simultaneously and need a tool to help streamline the planning process. Use iplan.ai to create, organize, and manage event details efficiently, ensuring no detail is overlooked.

How to access the tool:

  1. Sign up or log in to iplan.ai.

  2. Create a new event and enter the event details.

  3. Use the AI-powered suggestions to optimize your event planning.

  4. Access planning tools for scheduling, budgeting, and task management.

  5. Review and finalize your event plan with real-time updates.

Here is how you can utilize this application:

Use iplan.ai to streamline your event planning process, ensuring all details are efficiently managed and optimized for a successful event.

  • The Master Teacher of AI, Andrej Karpathy is back with an exciting tutorial. This time he explores how to build and optimize the GPT-2 model from scratch, enhancing your understanding of neural network training and enjoying the creative outputs it generates. Must watch, if you want to create intelligence hands-on.

  • In the recent episode of Leading with Data, I had a conversation with Dr. Honnibal of Spacy about his experience in developing cutting-edge language models and navigating the evolving landscape of AI technology.

  • This interview between Sam Altman and Nick Thompson focuses on leveraging AI to support global development goals such as health, climate, and sustainability, as discussed at the AI for Good Global Summit organized by the ITU and UN agencies.

  • Recently, Andrew Ng, founder of deeplearning.ai, shared a post about a pioneering approach to machine translation. He introduces an open-source project that utilizes an agentic workflow for language translation, which involves prompting an LLM, reflecting on the output, and refining it based on feedback. 

  • In a podcast by Dwarkesh Patel, Francois Chollet discussed the $1 million ARC-AGI Prize launch and shared his views on why large language models (LLMs) are unlikely to lead to artificial general intelligence (AGI).

What was your reaction to “Apple Intelligence”? Would love to hear it from you.

While the creator in me loved it, the techie in me was itching for more details! How would we as consumers know the evolution of Apple Intelligence? How would they announce updates to it? How is it performing on the popular benchmarks?

We will need to wait to see this play out and the techie in me is impatient, to say the least.

Until next week…

How do you rate this issue of AI Emergence?

Would love to hear your thoughts

Login or Subscribe to participate in polls.

Reply

or to participate.