04 Sep 2025

Google (Mostly) Wins; OpenAI Complies; Digital Twinsies

⚖ Google Avoids 'Breakup' but Faces New Restrictions

A federal judge ruled a big one - that Google will not be forced to spin off Chrome or Android in the search antitrust case, despite earlier speculation that competitors like OpenAI or Perplexity might swoop in if Chrome were cut loose. Judge Amit Mehta wrote that the rise of generative AI, including tools like ChatGPT, had already reshaped the competitive landscape in search.

While Google keeps its core businesses intact, it must stop signing exclusive default agreements with partners and share certain search data with competitors. Also Google can continue paying Apple and others for search placement as long as agreements aren't exclusive, preserving approximately $20 billion in annual payments. This means Chrome remains the cornerstone of Google’s ecosystem, but the regulatory spotlight on its distribution tactics will continue to burn bright.

Read more: The Rundown AI • The Verge

🍏 Apple Reportedly Testing Gemini for Siri Answers

Bloomberg reports that Apple is experimenting with running Google’s Gemini models inside its private cloud to power richer, multimodal “World Knowledge Answers” in Siri. If successful, this integration could go live in 2026, giving Siri a much-needed upgrade in factual responses and generative fluency.

The Rundown notes that Apple has also pulled back from acquisition talks with Perplexity, opting instead to license external models while building its own stack. Apple appears to be playing a balancing act—keeping control of data and UX while hedging with partnerships to avoid falling behind.

Read more: The Rundown AI • Bloomberg (paywalled)

👨‍👩‍👧 OpenAI Introduces Parental Controls for Teen Accounts

OpenAI announced new safety features for teenagers using ChatGPT. Guardians will be able to link accounts, set filters, and receive alerts when conversations suggest distress. OpenAI says it worked with clinicians to design these guardrails, and the rollout will happen within 30 days.

This is a major shift toward regulated AI adoption in younger demographics. With teens as one of ChatGPT’s fastest-growing user groups, OpenAI is laying down infrastructure to reassure regulators and parents alike that its tools are safe and supervised.

Read more: OpenAI • The Rundown AI

🔭 Perplexity Flies Comet Browser to Students

Perplexity announced that its Comet AI browser is now available to students, with PayPal helping to distribute early access. The move reflects a push to seed adoption in education and encourage younger users to shift from conventional search to AI-first browsing.

For Perplexity, this is both a distribution play and a bet that students will form habits around an AI-native browser, making Comet an everyday interface for knowledge and coursework.

Read more: Perplexity

🎙 “The VibeVoice" Battle is On!

AI Secret has an essay titled The VibeVoice War. The theme explores how real-time multimodal voice agents are colliding in product design and user experience, with OpenAI, Google, Kyutai, and smaller labs all racing to define what “speaking with AI” should feel like.

The essay hasn’t been posted publicly, but AI Secret frames this as a defining competitive battle for the next wave of assistants—where latency, personality, and safety tuning matter as much as accuracy. More if the story is posted.

Read more: AI Secret

🛠 Tools & launches

🎧 ElevenLabs Upgrades Sound Effects

ElevenLabs has released a new version of its sound-effects model. Clips can now run up to 30 seconds, support seamless looping, and generate higher-quality 48 kHz audio. The company also added new prompt controls to help creators dial in ambience and Foley effects. This pushes ElevenLabs closer to being a one-stop audio production suite, not just a TTS engine.

Read more: ElevenLabs docs

🗣 Respeecher Debuts Real-Time TTS

Respeecher launched an API for real-time streaming speech synthesis, boasting 200–300 ms latency and access to a marketplace of licensed voices. The low-latency target makes it viable for live agents, multiplayer games, and assistive devices, where responsiveness is critical.

Read more: Respeecher

🖥 Wispr Flow Adds IDE Awareness; Talk to Code

Wispr Flow, a voice-first coding assistant, now recognizes file names and variables when dictating code inside IDEs like Cursor and Windsurf. Developers can literally “talk through” code edits without breaking flow, and Wispr interprets both syntax and context. It’s a small but meaningful step toward hands-free, multimodal programming.

Read more: Wispr Flow

🎥 Tencent Releases HunyuanWorld-Voyager

Tencent open-sourced its Voyager model, which can generate RGB-D video conditioned on custom camera paths. The model enables long, coherent scene traversals and supports reconstruction into 3D worlds. By releasing weights and code, Tencent is encouraging academic and developer adoption in simulation and robotics research.

🧑‍💻 How to Build Marketing Videos with HeyGen Twins

The Rundown included a tutorial showing how to use HeyGen’s digital twin avatars to script, generate, and export marketing clips. This workflow demonstrates how execs can scale internal comms and external videos without needing a production team.

Read more: The Rundown AI

🗂 Everything Else Today

OpenAI Projects comes to free users - Free accounts now get project “workspaces,” with larger file-upload limits and project-level memory... Read more: Business Today
Alex coding assistant team joins OpenAI Codex - The indie Xcode assistant “Alex” will shut down new downloads in October; its team moves to OpenAI... Read more: Alex blog
Scale AI sues former exec and rival Mercor - Scale alleges a departing leader stole 100+ confidential documents and used them to poach clients. Mercor denies misuse... Read more: Axios • The Verge
NotebookLM expands audio formats — Google’s note-summarization tool adds new audio overview styles like “brief,” “critique,” and “debate,” plus more voices and video output options... Read more: Google blog