Apple Going AI Glasses, China Claims back the US AI Talent, and Anthropic Mythos Myths & Realities.

Show in Folder

MP’s Key Takes Today:

1. As I discussed last year, is leaning into mainstream AI Smart Glasses. Bloomberg now reports the product line up and plans. Apple’s entering Meta’s territory with the one edge Meta can’t match — 2+ billion iPhones to leverage the smart glasses.

2. China’s AI talent pipeline is flipping. Over 30 elite researchers left Silicon Valley for ByteDance, Tencent, and Alibaba in the past 12 months. Tsinghua’s US PhD applications dropped from 50% to 20%. The US is export-controlling chips while the people who know how to build the models are going home. It’s a growing issue for the US because China is half the world’s AI talent and market.

3. Anthropic’s new Claude Mythos is being launched in a swirl of publicity around cybersecurity concerns, and the reality of its capabilities. Mythos myth and reality. It will take a while to get a fuller picture. But for now, OpenAI’s is following the same cybersecurity playbook with its new ‘Spud’ model.

Apple Goes in on AI Glasses:

Bloomberg’s Mark Gurman broke the story: Apple is working on multiple smart-glasses frame styles with a unique camera design, codenamed N50. Four frame styles in testing — acetate material, black, ocean blue, light brown — including one that resembles Tim Cook’s own glasses. Vertically-oriented oval camera with surrounding lights. Unveil end 2026 / early 2027, release 2027.

The headline is that Apple’s building glasses. The bigger story is that Apple is now ready with its unique take on a category already crowded with existing players. From Meta and Snap today, to Google, Amazon, and possibly OpenAI/Jony Ive tomorrow.

Apple enters Meta’s territory two years late. Meta Ray-Ban has 3M+ active in market, 7M sold last year, targeting 20M by end of 2026 with the EssilorLuxottica partnership. That’s a real head start.

By its track record over fifty years, Apple isn’t really “late” — Apple enters markets late by design. The iPod pattern. The iPhone pattern. Ecosystem integration beats first-mover advantage when the ecosystem is 2 billion iPhones deep.

What N50 reportedly does. Photo and video capture. Phone calls. Notifications. Music. Upgraded Siri via iOS 27, powered by Apple Intelligence — which runs on Google Gemini underneath. Tethered to iPhone for editing, sharing, calls, and Siri. That tether is the point of leverage, as I explain here in today’s AI Ramblins #53.

The three-pronged wearables play. Smart glasses + camera pendant + enhanced AirPods, all feeding contextual awareness into Siri. Less social media, more personal assistant, across three form factors. Plus six other companies running at the same category: OpenAI + Jony Ive (Sweetpea, screen-free, 2027+), Google + Warby Parker + Samsung (Android XR), Amazon (Transformer / Alexa phone), Snap, Samsung. Six strategies, one post-smartphone form factor war, inning one.

MP angle: Been ‘in the rooms’ of every major tech wave since ‘82 at Goldman — seeing Apple entering a market “late” with ecosystem integration is their most proven playbook. The iPod was years behind Creative Technologies. The iPhone was years behind BlackBerry. But Meta has 7M Ray-Bans in the wild, EssilorLuxottica’s retail footprint, and a two-year data advantage from real-world usage. The 150+ million Americans who wear glasses daily are the addressable market both companies are racing toward. This is the form factor the post-smartphone era gets decided on, and Apple seems like they’re really in. (AI-RTZ #1055, #863, #1002, #849)

China Drains back AI Brains (After the US Trained Them)

The Financial Times and Implicator.ai report this week that elite AI researchers are leaving Silicon Valley for Chinese tech giants at an unprecedented rate. This is the human-capital dimension of US-China AI competition — and it may matter more than chip export controls.

The named returnees read like a who’s-who. Wu Yonghui: senior Google DeepMind researcher, now leading ByteDance’s Seed lab on next-gen LLMs. Yao Shunyu: 27-year-old ex-OpenAI researcher, joined Tencent on a 100 million yuan package (~$14M), reporting directly to Tencent president Martin Lau. Roger Jiang: left OpenAI to found a robotics startup in Shenzhen. Zhou Hao: Google DeepMind to Alibaba model refinement.

The pipeline stat that should worry Washington. Tsinghua engineering grads applying to US PhDs: dropped from ~50% pre-COVID to ~20% now. That’s the pipeline drying up at the source within a single political cycle. If it hits 15%, structural closure becomes irreversible.

The acceleration. Three headhunters (China + SF-based) placed 30+ US-based AI researchers in China in 12 months. Previous pace: low single digits per year. That’s an order-of-magnitude jump. Modern Diplomacy data adds another 85 established scientists moving from US to Chinese research institutions during 2025 alone.

The push-pull dynamics are now symmetric. Push from the US: rising H-1B visa fees, residual chill from the China Initiative (formally ended 2022 but the suspicion lingered), federal research funding cuts, and over 70% of surveyed Chinese-origin researchers reporting they feel academically insecure in US positions. Pull to China: pay now surpasses Silicon Valley adjusted for tax and cost of living. Tencent restructured its AI Lab under a “researcher-centered” model — no hard metrics, multi-year AGI programs, DeepMind-style template. ByteDance’s Doubao has 100M DAU. WeChat has 1.4B users. Shenzhen has 100+ humanoid robotics companies.

The closed-models twist. The talent returns AND the models go closed. Alibaba’s Qwen3.6-Plus and Qwen3.5-Omni: proprietary. Z.ai’s GLM-5-Turbo: closed. ByteDance Seedance 2.0: proprietary. The open-source crown, which moved East over the last 18 months, may be quietly shifting again. And in March 2026, Chinese authorities began urging top AI scientists to skip international conferences — a direct response to Meta’s acquisition of Manus, a Chinese-founded AI agent startup.

MP angle: This is the human-capital version of “you can’t export-control a model someone can download on GitHub.” The US has spent billions restricting chips. Meanwhile, 72% of researchers publishing at elite AI conferences from Alibaba, Huawei, and Tencent trained in North America. We trained them. They’re going home. And they’re taking the know-how of how to build frontier models with them. Chip controls can slow compute. The US government has done almost everything possible to put up ‘Chinese students/researchers go home’ signs. (AI-RTZ #767, #744, #819)

Anthropic’s Mythos: The Cybersecurity Tool They Won’t Release

Anthropic’s latest model Mythos has a swirl of myth and reality around it. Lot to unpack here over time.

MP angle: Anthropic’s playbook is now fully visible. Claude Code wins the developer market ($2.5B ARR, “Claude Mania” at HumanX). Mythos wins the security and government market (40+ critical infrastructure partners, Treasury in the loop). The Super Bowl “no ads ever” ad wins trust. OpenAI went media company with $2.5B in ad revenue. Anthropic went cybersecurity company with tools the government can’t ignore. That’s the fork from Ep 52, playing out in real time. (AI-RTZ #1051, #986)

Subscribe to AI-RTZ for the daily Substack and follow AI Ramblings Daily on YouTube for the video breakdown every morning.

Prefer shorter cuts? Here’s the buffet.

Not in the mood for the full 25-minute episode? Pick your depth. Four segment deep-dives and five bite-size shorts on the sharpest moments from today’s show.

🎯 Segment deep-dives (4-8 min each)

Apple’s Smart Glasses: Can the iPhone Ecosystem Beat Meta’s Two-Year Head Start? Apple seems a late entrant, but the market has barely gotten started. Meta has 3M+ Ray-Bans in market with a two-year head start. In a world with five plus billion folks with smartphones. Can Apple’s 2 billion-iPhone installed base close that gap? The iPod-over-Creative playbook, applied to wearables.

AI Market Share: The Big Five Racing for Distribution OpenAI leans consumer. Anthropic leans developers and enterprise. Google Gemini is deep in both camps. Microsoft leans in on AI Copilot. Meta is in 3.5+ B daily user habits. Reach beats benchmarks — and the distribution stage of AI is kicking in.

AI’s Early Adopter Divide: From Anthropic to OpenAI The AI adoption gap isn’t age or income — it’s curiosity. Early adopters are running circles around everyone else. The gap will look like the 1996 internet divide, except much faster.

Claude as a Coworker: The Real Pain Points of AI Early Adoption The ceiling isn’t the AI model — it’s your own preparation. Standing rulebook, attention capture system, nightly session log. Three to five days per habit you’re replacing. The shift from “chatbot I query” to “processes running for me” is where the labor savings actually land.

⚡ Quick hits (under 1 min each)

US Losing Its Top Chinese AI Researchers — And They’re Not Coming Back 30+ elite researchers left Silicon Valley for ByteDance, Tencent, and Alibaba in 12 months. Tsinghua US PhD applications cut in half. We trained them. They’re going home.

Apple’s Smart Glasses: 2+Billion iPhones Is the Edge Meta Can’t Match Meta has the head start. Apple has 2 billion iPhones as a distribution base. That’s the math nobody’s factoring into the AR glasses race.

Apple Entering Late Is Their Playbook — iPod & iPhone Classic Apple was late to MP3 players. Late to smartphones. Late works when your ecosystem is the moat. The N50 glasses are running the same playbook.

We’re in the MS-DOS Era of AI — And Most People Don’t Know It Yet Today’s AI interfaces feel advanced. Historically, they’re MS-DOS. Text in, text out, user does the hard work of knowing what to ask. We’re 3-5 years from the graphical revolution of AI.

AI Early Adopters vs Regular Users — The Gap Is Already Huge Curiosity is the divide. Early adopters have moved past chat into processes. Everyone else is still typing prompts. The gap compounds weekly.

MP’s Key Takes runs daily on AI Ramblings Daily — three differentiated calls on the day’s biggest AI news. Track them over time. See how many become conventional wisdom.

That’s it for today’s AI Ramblings Daily, #53. Stay tuned.

Follow MP on X/Twitter:

(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here)