- In The Loop AI
- Posts
- Google Can Train Search AI With Web Content Even After Opt-Out
Google Can Train Search AI With Web Content Even After Opt-Out
Apple Taps Anthropic to Power New AI Coding Tool
Algot Persson
May 02, 2025
In partnership with
In Today’s Issue:
Google Can Train Search AI With Web Content Even After Opt-Out
Apple Taps Anthropic to Power New AI Coding Tool
Read time: 3 minutes
Google Can Train Search AI With Web Content Even After Opt-Out
Google’s AI-powered search tools can be trained on web content even when publishers opt out, according to testimony from a company executive during a federal antitrust trial.
Eli Collins, VP at Google DeepMind, said Friday that the company’s opt-out mechanism only applies to DeepMind, not to Google’s broader search division. That means data excluded from DeepMind training can still be used to improve AI features like AI Overviews in Search.
“Once you take the Gemini model and put it inside the search org, the search org has the ability to train on the data that publishers had opted out of training, correct?” asked DOJ attorney Diana Aguilar.
“Correct — for use in search,” Collins responded.
The exchange came during a Washington trial to determine remedies after Google was found to have illegally monopolized search. The DOJ is asking the court to consider measures such as forcing Google to divest Chrome and blocking its default search deals on phones and browsers — including defaults that would benefit its AI models.
A document shown in court detailed how Google filtered out 80 billion of 160 billion training tokens from publisher opt-outs. Other data used for training included YouTube content and user “search session” logs.
Use AI as Your Personal Assistant
Ready to save precious time and let AI do the heavy lifting?
Save time and simplify your unique workflow with HubSpot’s highly anticipated AI Playbook—your guide to smarter processes and effortless productivity.
Apple Taps Anthropic to Power New AI Coding Tool
Apple is working with Anthropic to build a new AI-based code assistant powered by the Claude Sonnet model, as part of an internal overhaul of its Xcode development platform, according to people familiar with the matter.
The new tool, which is currently being used inside Apple but hasn’t been announced, is designed to write, edit, test, and debug code using natural language prompts. It’s part of a broader effort to speed up Apple’s software engineering processes and catch up to rivals in generative AI.
Anthropic’s Claude model is widely regarded for its programming capabilities. Apple’s tool includes a chat interface and can also simulate user interface testing — a labor-intensive task when done manually.
The company had previously attempted to build its own AI coding assistant, Swift Assist, but never launched it. Engineers internally criticized it for generating inaccurate code and slowing development. The Anthropic integration is a sign that Apple is now more willing to rely on external models.
If successful, the company may offer the tool to third-party developers, many of whom rely on Xcode to build apps for iOS and macOS. Apple is scheduled to hold its Worldwide Developers Conference on June 9.
How was today's newsletter? |
Reply