Sponsored by
Hello readers,
Welcome to the AI For All newsletter! Today, we’re talking about Apple’s belated Siri AI announcement at this year’s WWDC, redesigning industrial cameras for AI, and more!
AI in Action: Apple’s Making Privacy the Product
For years, Apple's approach to AI looked less like a strategy and more like a delay. Siri lagged, promised features slipped, and competitors built reputations for assistants that actually worked. At WWDC 2026 this week, Apple finally showed what it's been building — and the architecture underneath is as interesting as the features on top. The new Siri runs on a custom 1.2-trillion-parameter Google Gemini model, roughly eight times larger than anything Apple had built in-house. Complex queries route to Nvidia Blackwell chips on Google Cloud. It's a significant departure for a company that has long insisted on controlling every layer of its stack.
What Apple didn't surrender was the privacy framing. Those Blackwell chips run with hardware-based confidential computing enabled, which encrypts user input, model weights, and inference results inside GPU memory during computation. The cloud operator can't read the data in plaintext. Apple's contract with Google also bars Google from training future Gemini models on Siri queries, and simpler requests still run entirely on-device using Apple's own Foundation Models. The net effect is a system that sources capability from one of the world's most powerful AI labs while maintaining — credibly, structurally — that it isn't handing over your data to get there.
The broader question this raises is whether privacy-as-architecture can actually become a durable competitive advantage. Most AI development has treated user data as a resource to be harvested for model improvement; Apple is betting on a different model where capability and privacy aren't tradeoffs. With Siri now folding context from your emails, messages, photos, and calendar into every interaction, the stakes of that bet are higher than ever. A billion devices, all running a personal AI that knows your life. Whether users trust the architecture enough to lean into it is the real product question Apple is trying to answer.
🔥 Rapid Fire
Analysis: AI Is Slowing Down
“One of my sources has come forward and brought me a story that will possibly burst the AI bubble. If you’re wondering what the story is, know that it’s the information I’ve wanted for years, delivered as I have always wanted it. Imagine what the worst possible thing for me to get would be and you’re probably close. I expect it to be out in the next two weeks. I can guarantee you it’ll be worth it, and you’ll be stunned by what I report.”
Commentary: The Hater’s Guide to the AI Bubble 3.0
SoftBank’s attempt to get a $6 billion margin loan for OpenAI stalls
“The primary reason for the stalled loan lies in the fact that OpenAI’s valuation lacks a basis convincing enough for banks.”
OpenAI and Anthropic consider cutting prices for tokens
OpenAI and Anthropic cannot afford to do this
“Drastic price cuts could potentially erode the profit margins of both companies, which already lose billions of dollars.”
The End of Asset-Light: How AI Is Accelerating Technology Spending
“Further analysis shows that cash simply isn't keeping up. Bank of America estimates that these hyperscalers are positioned to consume roughly 90% of their operating cash flow on capex in 2026, up from 65% in 2025. Amazon is projected to post negative free cash flow of $17-$28 billion this year.”
Only 26% of companies have a comprehensive view of AI costs
KPMG report contains AI hallucinations on benefits of AI
“The report made numerous false claims about the use of AI by organizations including UBS, claiming the Swiss bank integrates AI agents across investment advisory, risk management, and compliance monitoring.”
S&P 500 upholds rules — blocks entry for SpaceX, OpenAI, and Anthropic
Crusoe pauses work on Wyoming data center after Google raises concerns
Satya Nadella finds nice way to tell employees to stop tokenmaxxing
Where’s the shovelware? Why AI coding claims don’t add up
Google is liable for false statements in AI Overviews rules court
Lawyers on both sides of case caught using AI — judge goes scorched earth
Stop Fine-Tuning Models You Don’t Need
Fine-tuning sounds like the answer until you factor in the cost, the data pipeline, and the six months before a bigger model makes yours obsolete. Most of the time, prompt engineering or better context gets you there. But sometimes it doesn't — and that's where things get interesting.
In this free night session, Aaron Gallant covers the real tradeoffs behind fine-tuning LLMs, from synthesizing training data with frontier models to running PEFT and QLoRA on constrained hardware. You'll learn when smaller, specialized models actually beat throwing money at a bigger one — and why data curation is the work nobody wants to talk about. Built for engineers who want to make the right call, not just the cool one.
Live and remote. Wednesday, June 3 at 5 PM CT. Register now.
📖 What We’re Reading
For decades, industrial cameras did one job: capture images. A human or a simple rule-based algorithm reviewed those images and made a decision. That model worked well enough when production lines moved slowly, and defect tolerance was forgiving. Today, neither of those conditions exists on any competitive factory floor.
AI vision has entered the picture, and it is not a minor upgrade. It changes what cameras need to do, how they need to be built, and what the hardware underneath them must support. If you are designing an embedded camera product for industrial IoT or evaluating one for deployment, understanding these shifts is no longer optional.
