🚀 RELEASES & LAUNCHES
Apple's watchOS 27.3 beta brings Audio Intelligence: Live Rewind and Siri Recap on Apple Watch
Apple's watchOS 27.3 developer beta introduces always-on Audio Intelligence features on Apple Watch, including Live Rewind and Siri Recap.- Aimed at Apple Watch Series 12 and Ultra 4 owners (S11 chip required) with iPhone 16 or later, excluding iPhone 16e.
- Audio Intelligence includes four mic-driven features: Sound Recognition, Shazam Music Recognition, Live Rewind (15-second text snippets via double-pressing the Digital Crown), and Siri Recap (daily conversation summaries that auto-delete after 7 days).
- Sound Recognition and Shazam are available now; Live Rewind and Siri Recap are in beta, English-only, and unavailable in the EU at launch.
- Privacy design: audio is processed in the S11 chip's Secure Exclave, never stored, with end-to-end encryption for saved summaries and no speaker attribution; Siri Recap summaries are generated via Apple Foundation Models on Private Cloud Compute.
- Series 12 starts at $399 and Ultra 4 at $799; the features matter for accessibility (sound alerts for deaf or hard-of-hearing users) and hands-free recall, but raise ambient-listening privacy questions.
Google DeepMind adds Cheminformatics Skills to Antigravity for drug discovery
Google DeepMind released Cheminformatics Skills for Google Antigravity that let an agent run drug discovery workflows in about an hour instead of days.- Aimed at computational and bench chemists doing drug discovery, compound search, and medicinal chemistry analysis.
- Cheminformatics Skills run on RDKit with connectors to ChEMBL and PubChem; the LLM orchestrates while deterministic chemistry tools execute molecular calculations.
- Demo: using real lung cancer TKIs (crizotinib, lorlatinib), the agent retrieved thousands of compounds, curated 568, ran matched molecular pair analysis, trained potency models, and proposed novel rigidified macrocycles with synthesis routes in ~1 hour, versus multiple days per manual iteration.
- Available via the Science plugin in Antigravity; Google is offering 10,000 scientists free access to its highest usage tier and the Gemini 4 Argon model.
- DeepMind frames outputs as testable hypotheses, not finished medicines; molecular handling stays in verified code due to zero error tolerance in chemistry.
🔬 RESEARCH
- 📄 Source for Library of Layers paper proposes adaptively reordering transformer layers at runtime — arxiv.org
📊 BENCHMARKS & EVALS
- Microsoft open-sources STATE-Bench, a memory-agnostic benchmark for AI agent reliability
Microsoft released STATE-Bench, an open-source benchmark measuring whether AI agents improve with memory on realistic enterprise tasks.- Aimed at agent developers, memory researchers, and platform builders evaluating agent memory systems in production-like settings.
- Covers 450 tasks across three domains — travel, customer support, and shopping — with sandbox databases, domain tools, simulated users, and deterministic success assertions.
- Reports four metrics: pass@1, pass^5 (success on all five runs), LLM-judged UX score (1-5), and cost per task; GPT-5.1 without memory reliably completes fewer than half of tasks, with only ~30% pass^5 in travel.
- Includes a pluggable 'bring your own memory' interface, task generation/auditing tools, and a public leaderboard; released under MIT license on GitHub.
- It addresses a gap left by retrieval-focused memory benchmarks by testing procedural, state-mutating workflows where agents typically fail on process, not recall.
💰 BUSINESS & FUNDING
OpenAI says ChatGPT search queries now equal 20% of Google's volume
OpenAI announced that ChatGPT's search query volume now equals 20% of Google's.- For ChatGPT users, SEO/marketing professionals, and search-industry watchers tracking AI vs Google competition.
- Adam Fry, OpenAI's consumer product lead for ChatGPT, said at a Monday press event that ChatGPT's search query volume equals 20% of Google's.
- Independent analyses give lower estimates: Ahrefs puts ChatGPT at about 12% of Google's search volume (2.5B daily prompts vs Google's 13.7B searches), and First Page Sage pegs ChatGPT at 19% of digital queries in October 2026 vs Google's 75.7%.
- Traffic gap remains huge: Google sends 190x more website traffic than ChatGPT, with Google driving 39.98% of referral traffic vs ChatGPT's 0.21%.
- ChatGPT has already surpassed Bing (~1.2B daily searches), and AI search growth is shifting how builders should think about SEO and referral traffic.
Anthropic's compute margins are worsening even as scale grows
Anthropic's compute margins are deteriorating at a point when they should be improving, per TechCrunch.- Anthropic faces worsening compute economics despite rising scale, making it less like traditional software and more like a capital-intensive infrastructure business.
- The company carries massive compute commitments: $200B to Google Cloud, $100B to AWS, $30B to Microsoft Azure, and a ~$40B lease (about $1.25B/month through May 2029) on xAI's Colossus 1 data center with 220,000+ Nvidia GPUs.
- Anthropic reported $65B Series H funding at a $965B post-money valuation and a ~$47B annualized revenue run rate as of May 2026.
- Every token generated carries direct compute cost, breaking the software 'law of large numbers' that normally lets margins expand with user growth.
- For builders and adopters, Anthropic's economics depend on managing a tri-platform silicon strategy across Nvidia GPUs, Google TPUs, and Amazon Trainium.
Sources 📎 Coverage: techcrunch.com · 👁 via TechCrunch AIApple posts record fiscal year as earnings show AI strategy leaning on partners, not capex
Apple's latest earnings show record revenue and EPS while it pursues a low-capex AI strategy built on on-device AI and licensed models like Google's Gemini.- For Apple investors and iPhone users weighing the company's AI approach against big-tech rivals.
- Apple posted its best-ever fiscal year for revenue and EPS; Q4 iPhone revenue rose 17.5%, and one quarter showed adjusted EPS of $2.01 vs. $1.96 estimated on revenue of $111.2 billion.
- Apple keeps quarterly AI capex around $2 billion (December-quarter capex of $2.37 billion), versus Microsoft's $116 billion fiscal-2026 capex, up 80%.
- Apple reportedly pays Alphabet about $1 billion a year to power a revamped Siri with Gemini, while relying on an installed base of over 2.5 billion active devices for AI distribution.
- The results fuel debate over whether Apple's wait-and-see, partner-reliant AI strategy is a masterstroke or a vulnerability, with shares down roughly 10% since the last earnings report.
Sources 📎 Coverage: theverge.com · 👁 via The Verge AI
🏛️ POLICY & REGULATION
Google will fully disable Search in schools that opt out of its AI features
Google told schools that rejecting its AI features means losing Google Search entirely.- For K-12 schools and student accounts choosing to restrict AI altogether
- Google said access to Gemini, Gemini Notebook, Gemini in Classroom, and Google Search 'will be entirely off' for K-12 student accounts that reject AI features
- Because AI is now integrated into Search, schools can no longer disable AI while keeping the search engine
- Google also says it will offer schools AI tools that purportedly don't give away homework answers
- The move follows a Common Sense Media review that rated Google's AI search tools 'Unacceptable' for students, citing homework answers, inaccuracies, and missed safety red flags
Sources 📎 Coverage: techcrunch.com · 👁 via The Verge AI, TechCrunch AI (+1)Apple requires 18+ ratings for dating apps and AI romantic companions
Apple is adding App Store content descriptors that force dating apps and AI companions with sexual or romantic content to carry an 18+ age rating.- Affects dating apps and AI companion apps like Character.AI and Replika, plus any app sharing users' real-world location with others.
- Starting November 16, developers must answer new descriptor questions in App Store Connect; all apps must have declarations filed by January 12, 2027.
- Apps flagged for dating 'between real people and/or a virtual companion that engages a user in sexual or romantic content' get a minimum 18+ rating and are hidden from minors under parental content restrictions.
- Top AI companion apps had been downloaded 220 million times globally as of last summer; xAI also launched NSFW-capable 3D AI companions.
- Location-sharing won't change age ratings in most regions, though Brazil requires A16 for location-sharing and A18 for dating apps, with possible MJSP exceptions.