Jay Yagnik, VP of AI innovation and research, on Google’s The Keyword blog:

Private AI Compute is built on a multi-layered system that is designed from the ground up around core security and privacy principles:

• One integrated Google tech stack: Private AI Compute runs on one seamless Google stack powered by our own custom Tensor Processing Units (TPUs). World-class privacy and security is integrated into this architecture with Titanium Intelligence Enclaves (TIE). This design enables Google AI features to use our most capable and intelligent Gemini models in the cloud, with our high standards for privacy and the same in-house computing infrastructure you already rely on for Gmail and Search.
• No access: Remote attestation and encryption are used to connect your device to the hardware-secured sealed cloud environment, allowing Gemini models to securely process your data within a specialized, protected space. This ensures sensitive data processed by Private AI Compute remains accessible only to you and no one else, not even Google.

Sounds a lot like Apple’s Private Cloud Compute, which raises the question of whether this Google project is related to the Gurman scoop that Apple and Google are on the cusp of a deal for a white-label version of Google Gemini to run on Apple’s Private Cloud Compute servers to power the next-generation versions of Siri and Apple Intelligence.

I strongly suspect this is something Google has been working on for a while. Apple, I think it’s fair to say, places a higher priority on privacy than does Google, but Google does value privacy. But perhaps the deal with Apple accelerated the project within Google.

OpenAI:

To start a group chat tap the people icon in the top right corner of any new or existing chat. When you add someone to an existing chat, ChatGPT creates a copy of your conversation as a new group chat so your original conversation stays separate. You can invite others directly by sharing a link with one to twenty people, and anyone in the group can share that link to bring others in. When you join or create your first group chat, you’ll be asked to set up a short profile with your name, username, and photo so everyone knows who’s in the conversation. Group chats can be found in a new clearly-labeled section of the sidebar for easy access. [...]

Group chats are separate from your private conversations. Your personal ChatGPT memory is not used in group chats, and ChatGPT does not create new memories from these conversations. We’re exploring offering more granular controls in the future so you can choose if and how ChatGPT uses memory with group chats.

Currently rolling out in Japan, New Zealand, South Korea, and Taiwan. [Update: Expanded worldwide just three days later.] Rate limits and advanced model usage depend on the group member to whom ChatGPT replies. Pretty clever, and I can imaging a lot of ways this would be useful, both for family/friends and for work collaboration. I like the idea that this is built into ChatGPT, not an AI bot in a regular messaging app. This way, you know with certainty which of your chats are being seen and read by an AI bot.

We’ve also taught ChatGPT new social behaviors for group chats. It follows the flow of the conversation and decides when to respond and when to stay quiet based on the context of the group conversation. You can always mention “ChatGPT” in a message when you want it to respond. We’ve also given ChatGPT the ability to react to messages with emojis, and reference profile photos — so it can, for example, use group members’ photos when asked to create fun personalized images within that group conversation.

This is a really hard problem to solve. Wavelength, the late great private group messaging app whose team I advised from 2023 to 2024 (when the app shuttered), added AI chatbots (with customizable personalities) in June 2023. Wavelength’s AI bots only responded when mentioned explicitly.

My thanks to WorkOS for their continuing support of DF with another sponsorship week. With WorkOS you can start selling to enterprises with just a few lines of code. WorkOS provides a complete user management solution along with SSO, SCIM, and RBAC. The APIs are modular and easy-to-use, allowing integrations to be completed in minutes instead of months. WorkOS simplifies MCP authorization with a single API built on five OAuth standards.

Today, over 1,000 companies rely on WorkOS, including large fast-growing startups like OpenAI, Cursor, and Vercel. And for companies just getting started, WorkOS offers up to one million monthly active users free of charge.

For SaaS apps that care deeply about design and user experience, WorkOS is the perfect fit. From high-quality documentation to self-serve onboarding for your customers, it removes all the unnecessary complexity for your engineering team.

The Financial Times, under a four-person byline (“Tim Bradshaw, Stephen Morris, and Michael Acton in San Francisco, and Daniel Thomas in London”):

Apple is stepping up its succession planning efforts, as it prepares for Tim Cook to step down as chief executive as soon as next year. Several people familiar with discussions inside the tech group told the Financial Times that its board and senior executives have recently intensified preparations for Cook to hand over the reins at the $4tn company after more than 14 years.

John Ternus, Apple’s senior vice-president of hardware engineering, is widely seen as Cook’s most likely successor, although no final decisions have been made, these people said.

People close to Apple say the long-planned transition is not related to the company’s current performance, ahead of what is expected to be a blockbuster end-of-year sales period for the iPhone. [...]

The company is unlikely to name a new CEO before its next earnings report in late January, which covers the critical holiday period. An announcement early in the year would give its new leadership team time to settle in ahead of its big annual keynote events, its developer conference in June and its iPhone launch in September, the people said. These people said that although preparations have intensified, the timing of any announcement could change.

I have no little-birdie insight on this, but that’s not surprising. I don’t think there are many people, if any, outside Apple’s top executive team and board of directors who have any insight into Cook’s thinking on this. That “several people” spoke to the FT about this says to me that those sources (members of the board?) did so with Cook’s blessing, and they want this announcement to be no more than a little surprising.

I absolutely love the idea of Cook’s successor being a product person like Ternus, and Ternus is young enough — 50, the same age Cook was in 2011 when he took the reins from Steve Jobs — to hold the job for a long stretch. Ternus took over iPhone hardware engineering in 2020, and was promoted to senior vice president of hardware engineering in January 2021, when Dan Riccio stepped aside. Apple’s hardware, across all product lines and including silicon, has been exemplary under Ternus’s leadership. And Ternus clearly loves and understands the Mac.

I would also bet that Cook moves into the role of executive chairman, and will still play a significant, if not leading, role for the company when it comes to domestic and international politics. Especially with regard to Trump.

The question is, did Meta scrap its native Windows app because they don’t care that much about Windows in particular? Or because they don’t care that much about native desktop apps, period — and a crude web app wrapper is coming to Mac next?