Models & Labs
Apple Watch Audio Intelligence Beta Begins
TechCrunch AIhigh confidence
Why it matters
- →Ambient conversation recording moves from concept to consumer beta on mainstream hardware.
- →Privacy architecture uses secure enclaves and ephemeral cloud processing to mitigate surveillance concerns.
- →Sets a precedent for how wearable AI handles sensitive audio data in public spaces.
Apple has begun beta testing its 'Audio Intelligence' features for the Apple Watch Series 12 and Ultra 4 with iOS 27.3 developers. The suite includes Live Rewind, which captures the last 15 seconds of conversation as text, and Siri Recaps, which generates summarized notes from ongoing dialogues. Processing relies on the S11 chip to detect speech, a secure enclave to buffer audio, and Apple's Private Cloud Compute for AI summarization. Raw audio is not stored, and summaries are automatically deleted if not saved within seven days.
Read originalThe story around this
TopicApple Siri AIRising
Earlier coverage that leads up to this article, and what followed. Lines connect each piece to the closest one after it, converging here.
How we got here
- 1Revamped Siri AI Changes iPhone Interaction
The Verge AI · July 13, 2026 · Background
- 2Apple's Siri AI Revamp Debuts in iOS 27 Beta
WIRED AI · July 13, 2026 · Background
- 3Apple's Siri AI Opens to Public with iOS 27 Beta
TechCrunch AI · July 14, 2026 · Related
- 4MemoketAI Launches Gem, a Wearable AI Assistant
Matt Wolfe · August 19, 2026 · Background
- 5Apple Watch's AI Features Raise Privacy Concerns
TechCrunch AI · September 9, 2026 · Same story
- 6Apple Unveils Privacy-Focused Siri Audio Features
The Verge AI · September 9, 2026 · Same story
- 7Apple's Siri AI Gets Major Upgrade with iOS 27
WIRED AI · September 10, 2026 · Background
More from TechCrunch AI
© TechCrunch AIGeneral AIother
Google expands Dreambeans AI stories to India
Google’s experimental personal story app Dreambeans has left the U.S., launching in India with support for 13 local languages. The app pulls data from Gmail, Photos, and Calendar to generate illustrated narratives, recently adding Veo-powered animations. Despite high-profile competitors like Meta’s Muse, Dreambeans saw modest U.S. adoption, raising questions about its mass-market appeal. The expansion tests whether personalized AI memory tools can scale in a multilingual market where code-switching is the norm.
TechCrunch AI
© TechCrunch AIMarket & Regulationbusiness
Apple eyes third-party AI app revenue
Apple is quietly pivoting to monetize the AI wave through its App Store, signaling that third-party AI agents could become a new revenue stream. This shift addresses regulatory headwinds and an 18% drop in U.S. commission revenue by potentially taking a cut of AI agent transactions. While details remain vague, the move suggests Apple sees AI as a key driver for upgrades and subscription growth. It marks a strategic adaptation to maintain services dominance amid global legal pressures.
TechCrunch AI
© TechCrunch AIInvestment · $45M
Market & Regulationbusiness
Trajectory raises $45M for automated AI continual learning
Trajectory’s $45 million Series A signals a pivot toward making post-training accessible to non-technical experts. By automating continual learning, the startup lets domain specialists like lawyers and doctors refine open-source models through feedback loops without hiring expensive research teams. This approach bridges the gap between foundational models and specialized industry needs, potentially lowering the barrier for enterprise AI adoption. With backing from Sequoia and Nvidia, Trajectory is positioning itself as a critical infrastructure layer for vertical-specific AI applications.
TechCrunch AI
More in Models & Labs
Models & Labscoding
vLLM v0.33.0: Blackwell & DeepSeek V4.1 Optimizations
vLLM v0.33.0 is a massive infrastructure update that finally brings first-class support for NVIDIA's Blackwell architecture and the new DeepSeek V4.1 models. The release drops CUDA 12.9 in favor of CUDA 13.0, signaling a hard pivot to next-gen hardware while adding critical optimizations like block-diffusion drafters and quantized LM heads that significantly boost decode throughput. AMD users also get major wins with ROCm 10.0 defaults and MonoKernel decoding on MI355X. This isn't just another version bump; it's the runtime catching up to the bleeding edge of model complexity and silicon capability.
vLLM Releases
Models & Labsother
llama.cpp b12054 adds CUDA 13 and ROCm 10.0
This release quietly cements llama.cpp as the universal inference runtime by adding default support for CUDA 13 and ROCm 10.0. NVIDIA users can now leverage newer driver stacks without waiting for separate builds, while AMD GPU owners finally get parity with modern ROCm versions. The inclusion of Snapdragon and OpenVINO binaries further widens the hardware net, though Apple Silicon KleidiAI is notably disabled here. It’s less about new features and more about ensuring no major accelerator is left behind in the local LLM ecosystem.
llama.cpp Releases
Models & Labsmodels
llama.cpp adds Clover MoE model support
llama.cpp quietly expands its architecture roster by adding native support for the Clover Mixture of Experts model. This isn't just a wrapper; it implements causal hybrid attention and reuses existing recurrent-state and LoRA helpers, meaning local inference for this specific architecture is now viable on consumer hardware. The release includes GGUF conversion tools and loads across all major backends including CUDA 13, ROCm 10, and Apple Silicon. For researchers tracking non-transformer or hybrid architectures, this closes a gap in the most popular local inference runtime.
llama.cpp Releases