Hey Devs!

Welcome to the AI for Developers Newsletter—where the AI buzz is hotter than your CPU after an all-night coding session! We’re here to inject some life into your inbox with juicy insider tidbits from the AI dev grapevine.

Let’s dive in and ‘feel the AGI!’ ;)

Just when we thought Apple was asleep at the AI wheel, they burst onto the track at WWDC 2024, announcing a tidal wave of AI features that reset the race for the competition.

‘Personal Intelligence for the Masses’

Apple has ingeniously branded its AI push as "Apple Intelligence" – personal intelligence that understands you. For developers, the key lies in Apple's agent-based foundational model layered on top of the OS, with different training passes to teach specialised capabilities.

Siri now calls LLM functions at the system app level, allowing users to work seamlessly without grasping the underlying tech.

The Developer Lowdown

• On-device intelligence stack with a compact, low-latency foundational language model for system integrations.
• Semantic index for understanding on-device context.
• Specialised sub-models ("Adaptors") for tasks like text summarisation, tone adjustment, proofreading, and more.
• Compressed, optimised model outputs for mobile efficiency.
• Diffusion model with "Adaptors" for creating and modifying images.
• "App Intents" APIs for Siri to recognise app functions and integrate with voice/text commands.

Privacy First, Compute Second

True to form, Apple processes as much as possible on-device before parsing to the cloud (with user permissions). Despite partnering with OpenAI to offer free ChatGPT access, Apple maintains its privacy stance by employing the following:

• Hardened iOS subset with no persistent storage and tooling to prevent remote access
• Full machine learning stack
• Secure enclaves to protect device encryption keys
• Two-way verification— both device and cloud verify identity before sending requests
• Cloud cluster decrypts data and generates response but doesn't retain it

Apple has promised to publish production images of its private cloud system for public inspection, and each cluster will be part of a publicly signed and verified build.

Apple & OpenAI, Apple’s First AI Team Up

Apple's first public AI partner is OpenAI's ChatGPT, which users can access for free (paid subscriptions are also supported). Siri can tap into ChatGPT’s answers and serve them to the user, and ChatGPT will also be available in the Apple Writing Tools system for text generation.

Notably, Apple did not commit to OpenAI, revealing later that it plans to partner with Google to use Gemini and leaving open the possibility that it might also add Anthropic for Claude.  

Development Opportunities Unlocked

For developers, a whole new set of opportunities has materialised. With Apple’s marketing prowess, the future of AI development looks brighter than ever. AI-powered features, enhanced developer tools, and the integration of generative AI open doors for innovation and expanded capabilities across Apple's ecosystem. For a fuller breakdown of everything Apple Intelligence-related, check out our blog post and visit the Apple Developer site for more details.

NVidia: Bigger than Ever

Nvidia CEO Jensen Huang has been bullish on the idea that generative computing represents a trillion-dollar opportunity. With Nvidea becoming the world’s second-most valuable company, it’s easy to see why he might say that. Computex 2024 represented another significant opportunity for the company to make significant announcements aimed at AI developers.

Key Announcements:

Customise, Optimise, Deploy: Developers can now supercharge their genAI models on RTX AI PCs. Customise with QLoRa and optimise with TensorRT for 4x faster deployment in packages 1/3 the size.

Nvidia NIM: AI in a Box Want to add text, image, video or speech genAI to your app? Nvidia's optimised model containers make deployment a breeze with their new NIM (Nvidia Inference Microservices) platform. 

Generative Goodies Other highlights include Project G-Assist for AI game assistants, digital human NIMs, and a collaboration with Microsoft for on-device Windows Copilot AI integration using GPU-accelerated small language models.

Nvidia is leaving no stone unturned in the generative AI gold rush. Developers, get ready to mine those trillion-dollar opportunities. (Link)

Mistral: The French AI Startup That Thinks It Can Beat Meta and Musk

Do you think you can't outsmart tech giants like Meta and Musk? Well, French startup Mistral begs to differ. Mistral has blown up to a valuation of €2 billion in less than a year, joining an exclusive club of French tech unicorns. 

With its customisable AI models, LoRA-based SDK, and serverless fine-tuning services, Mistral is betting on allowing developers to tailor AI models to their needs. 

Bespoke AI tailored to your exact needs? Mistral thinks it's worth a shot and is cutting its code to measure. 

Is this ambitious startup just wasting thread, or do they have a shot at disrupting the AI landscape? Find out by clicking the link and decide for yourself.

FineWeb: The Michelin-Starred Dataset for Your AI Appetite

Forget dirty data! Hugging Face, the cool kids of NLP, just dropped FineWeb, a squeaky-clean dataset that will make your language models smarter than ever. Imagine 15 trillion tokens of pure, meticulously filtered web goodness, ready to supercharge your AI projects.

Billing itself as ‘the finest collection of data the web has to offer,’ Hugging Face aims to reach data connoisseurs who understand the value of a cooked dataset and have the discerning palates required to appreciate the meal.   Read more.

Eric Schmidt's AI Crystal Ball

Former Google CEO Eric Schmidt is painting a vivid picture of how AI is about to reshape our world—and it's arriving faster than even a former tech CEO turned AI oracle can predict. From AI agents conspiring like a digital mafia to medical breakthroughs we could only dream of. The possibilities are equally thrilling and terrifying. But are we ready? Dive into Schmidt's mind-blowing predictions and decide for yourself. Click the link for the full scoop.

Snowflake Summit 2024: The Good, the Bad & the AI-Powered

Snowflake, the cloud data platform everyone's talking about, just wrapped its annual summit, unveiling a suite of AI-powered goodies for developers. But it's not all sunshine and rainbows. Amid the excitement of new tools for building models and managing data pipelines, dark clouds are gathering – a series of cyberattacks targeting Snowflake's customers have raised serious questions about the platform's security.

Is Snowflake's data fortress secure enough to withstand the marauders? Can its shiny tools mitigate security concerns? How does Snowflake compare to its cloud rivals regarding features and performance? Find out here.

Level Up Your Prompt Game: Part 3 of Our Prompt Engineering Series is Here!

In part 3 of our series, we're ditching the idea of one-and-done prompts and diving into the ones and zeros of iterative prompt development—where trial and error is not just okay, it's essential!

Click here to master the art of refining your AI prompts.

Until next time, may your bugs be few and your breakthroughs be many!

The AI for Developers Team