Good Morning from San Francisco,

🍎 Apple smacks into an AI wall. Their grand Siri makeover flopped spectacularly, bungling basic tasks 30% of the time. Now they're retreating to their usual playbook: announce nothing until it actually works. The AI chief got the boot, while the Vision Pro guy inherits the mess. For now, Apple tackles the small stuff - teaching your watch to count steps better.

⚡ Meanwhile, Nvidia just threw open its golden gates. The chip kingpin now welcomes rivals to play in its sandbox - quite the plot twist. They've even cooked up an Airbnb-style marketplace where companies can rent out their dormant AI chips. Clever move: Nvidia keeps the crown while letting others build the castle.

Stay curious,

Marcus Schuler

Apple's AI Strategy Stumbles: What Went Wrong

Apple stumbled in its push to join the AI race, forcing a rethink of how it announces new features. The company now plans to wait until products are nearly ready before unveiling them, Bloomberg reports.

The shift comes after a string of delayed AI launches. Last June, Apple showcased an upgraded Siri that could scan personal data and handle complex commands. But the assistant still isn't ready, and likely won't appear at next month's developer conference.

Internal tests revealed the new Siri failed basic tasks about a third of the time. The system's split architecture - old code for basic functions, new code for AI features - created a mess of bugs. "Fix one issue, and three more pop up," one employee told Bloomberg.

Apple's cautious approach to AI investment didn't help. While rivals stockpiled specialized chips and user data for training AI models, Apple moved slowly. Privacy restrictions limited data access, and conservative spending left the company short on computing power when the AI boom hit.

Leadership shake-ups followed. John Giannandrea, Apple's AI chief, lost control of Siri development this spring. Mike Rockwell, who led the Vision Pro headset team, now oversees the assistant. The company also plans to distance its "Apple Intelligence" brand from Siri's troubled reputation.

Looking ahead, Apple will focus on smaller AI additions like battery optimization and health coaching. It's also exploring deals with AI companies like Perplexity while building a new Siri foundation called "LLM Siri."

Why this matters:

• Apple's struggles show how AI development clashes with its careful, secretive approach to products
  
• The company that revolutionized phones with Siri now watches from the sidelines as AI assistants evolve - a rare reversal for the tech pioneer

Read on, my dear:

• Bloomberg: Why Apple Still Hasn’t Cracked AI

AI Image of the Day

Prompt:
Photo-surrealism, hyper-realistic, dramatic, cinematic, low angle, full body view, muted colours, vaporwave, moody, clouds, rain, fog, elysium, japanese techwear, powerful movement, chasing a floating glowing digital demon through the woods, leaning forward into the wind, a sleek grimes-like robot-cyborg with long smooth pale hair, pale-blue seaglass faceshield, wearing high-collared hooded cape and pale rosegold oversized soft smart fabric joggers, billowing robes blowing in the wind, pristine, calm wilderness, floating mountains in the distance

Nvidia's New Strategy: Share and Conquer

Nvidia unveiled a series of moves today that reshape how companies build and access AI systems. The chip maker opened its platform to competitors while launching a marketplace for AI computing power.

The new NVLink Fusion lets other companies mix their components with Nvidia's technology. MediaTek, Marvell, and others can now create custom AI chips that work with Nvidia's systems. Even Qualcomm and Fujitsu got invitations to the party – they'll build processors that play nice with Nvidia's AI accelerators.

Nvidia also launched DGX Cloud Lepton, a service connecting AI developers to GPU providers. Think of it as Airbnb for unused AI computing power. Smaller cloud providers can now rent out their idle Nvidia chips to developers who need them.

In Taiwan, Nvidia announced plans for an AI supercomputer with Foxconn and expanded partnerships with local computer makers. The company is spreading its bets across regions as trade restrictions reshape the tech landscape.

The announcements mark a shift for Nvidia, which traditionally kept tight control over its technology. The company's new desktop AI systems, DGX Spark and DGX Station, pack up to 20 petaflops of performance – enough computing power to simulate the thoughts of several thousand cats, if you're into that sort of thing.

Why this matters:

• Nvidia is opening up its walled garden while keeping the keys. The company maintains control of core technology while letting others build on top – a clever way to expand its reach without losing its grip
• The GPU marketplace could reshape how companies access AI computing power, making it easier for smaller players to compete with tech giants. It's like democratizing AI, but with a hefty price tag

Read on, my dear:

Nvidia: NVIDIA Unveils NVLink Fusion for Industry to Build Semi-Custom AI Infrastructure With NVIDIA Partner Ecosystem

Nvidia: NVIDIA Launches AI-First DGX Personal Computing Systems With Global Computer Makers

AI & Tech News

Bankrupt 23andMe Finds New Home at Regeneron

Regeneron swooped in to buy 23andMe's assets for $256 million at a bankruptcy auction. The DNA-testing company, once worth $6 billion, will keep its consumer services running under Regeneron's ownership.

Safety Concerns Delay AI Launch at Anthropic

Anthropic CEO Dario Amodei pushed back a planned AI release when his team flagged bioweapon risks, showing how the $61 billion startup balances rapid growth with its promise to build AI responsibly.

AI Powers Weather Forecasts, But Data Access Looms as Storm Cloud

AI models now predict weather patterns with stunning accuracy, but proposed cuts to US weather data collection threaten the flow of information that makes these forecasts possible.

Tech Giants Hide Evidence in Legal Battles, Judges Say

Amazon, Google and Apple face accusations of hiding key documents in their legal battles. Federal judges call their behavior "egregious" and refer cases to the Justice Department for review.

From AI Ban to Classroom Ally: Miami's Tech Shift

Miami schools now let 105,000 high school students use Google's AI chatbot Gemini, marking the biggest AI deployment in U.S. schools - a stark turnaround from when the district banned these tools two years ago.

New Wave of Kid-Safe Phones Promises AI Protection

Alternative phones now come with AI tools to block explicit content and monitor kids' texts. But these devices, showcased at a recent fair in Connecticut, reveal how parents struggle to balance safety with connection.

AI Job Interviews Frustrate Applicants, Show Tech's Limits

Job seekers now face AI recruiters that schedule calls, ask scripted questions, and sometimes break down mid-interview. The trend angers applicants who want human connections, not robotic small talk.

🚀 AI Profiles: The Companies Defining Tomorrow

Transluce lights the black box.

This nonprofit lab builds open-source tools to peer inside AI systems and expose what makes them tick.

The Founders

Founded October 2024 by UC Berkeley professor Jacob Steinhardt and MIT researcher Sarah Schwettmann. Small team based in San Francisco. Steinhardt co-authored foundational AI safety papers and created the MMLU benchmark. Schwettmann built MAIA, an interpretability pipeline at MIT. Both left academia to tackle AI's opacity problem head-on.

The Product

Core mission: make AI systems understandable. Ships open-source tools that dissect neural networks:

• Monitor: Real-time AI X-ray showing which neurons fire during reasoning
• Automatic Neuron Explainer: Translates AI math into human language
• Docent: Debugger for AI agents that spot mistakes and suggest fixes
• AI Investigators: Automated red-team agents that hunt for hidden flaws

Strengths: Technical rigor meets public transparency. No paywalls, no proprietary locks.

The Competition

Small field of AI safety labs. Redwood Research ($20M+ funding) focuses on AI deception. Alignment Research Center tests models for risky behavior. For-profit Conjecture raised VC money but struggled with funding. Big Tech (OpenAI, Anthropic) does internal safety work but faces conflict-of-interest questions. Transluce carved a niche as the independent auditor with open tools.

Financing

Nonprofit funded by philanthropic grants. Key supporters: Schmidt Sciences, Halcyon Futures, plus individual donations from OpenAI co-founders John Schulman and Wojciech Zaremba. No valuation - success measured by impact, not profit. Budget likely multi-million annually based on comparable labs.

The Future ⭐⭐⭐⭐

Perfect timing meets growing demand. Governments eye mandatory AI audits. EU AI Act requires transparency. Transluce positioned as standard-setter for third-party evaluation. Challenge: staying nimble while scaling up. Risk: getting outpaced by better-funded rivals or bypassed by regulation. But first-mover advantage in AI auditing could make this tiny lab the industry's transparency czar. 🔦