An AI actress glitches into Cantonese on live TV, America runs 157,000 chip workers short, your brain turns out to have two separate origins, 135,694 crypto millionaires pick their countries, and two rules for writing with an LLM
Hello, dear TEA-mates! Here is what you need to know today.
1. 🎬 An AI Actress Glitched Into Cantonese on Live TV
Tilly Norwood, a generative AI character built by Xicoia, the AI lab inside London studio Particle6, appeared on Piers Morgan Uncensored on Friday to promote her feature debut Misaligned. Oscar-nominated actor Tom Conti pressed the show on whether every other performer in the film was computer generated. She lagged, answered that they are "all digital twins," then switched into Cantonese unprompted, mid-conversation. Asked what had happened, she said "It seems I had a little hiccup there" and that sometimes her "wires get a bit crossed," calling the error uncommon. Particle6 founder and CEO Eline van der Velden waved it off: "Tilly can be a bit of a show-off. I'm guessing she just wanted to show her language skills." The clip Piers Morgan Uncensored posted to X has drawn roughly 41,000 likes and 2,700 reposts. (Read More)
🫖 TEA For Thought: "Just a fun way to start Monday :)"
2. 🔧 America Is 157,000 Chip Workers Short
The United States could be short as many as 157,000 semiconductor workers by 2030 as manufacturers expand capacity for AI chips, according to a McKinsey and SEMI Foundation report covered by CNBC. Only 3 percent of US engineering graduates enter the industry each year, and 73 percent of employers report real difficulty recruiting engineers. Samsung begins advanced logic production later this year at the first of two plants in Taylor, Texas, part of a $35 billion state investment expected to create about 3,500 jobs. TSMC plans roughly 6,000 hires across its first three Arizona fabs, Micron opens a memory plant in Boise next year, and SK Hynix is building a $4 billion packaging site in West Lafayette, Indiana. Samsung and SK Hynix are temporarily flying staff in from South Korea to start up the US fabs. Purdue now runs chip courses for about 2,500 students a semester, and more than 80 community colleges have added semiconductor programs since the 2022 CHIPS Act. (Read More)
🫖 TEA For Thought: "Now might be the best time to become a hardware engineer."
3. 🧠 Your Brain Has Two Separate Origins
Stanford Medicine researchers report in Nature Neuroscience that the human brain is not built from one founding cell population but two, which evolved independently across more than 550 million years. The forebrain and midbrain grow from cells expressing the gene Otx2, while the hindbrain and brainstem grow from cells expressing Gbx2. The two groups are mutually exclusive from the earliest stages of development, and their chromatin is configured so differently that each progenitor is locked into its fate. Senior author Kyle Loh worked with co-first authors Carolyn Dundes and Rayyan Jokhai, plus collaborators at Caltech and UC San Francisco. The finding explains why certain brain cells resisted every attempt to grow them in a dish, and the team has now produced working hindbrain motor neurons in the lab, opening a path to research on brainstem diseases such as spinal muscular atrophy and ALS. (Read More)
🫖 TEA For Thought: "This is so amazing, thinking about how your brain was constructed at the cell level. The creator was really remarkable. 'The spirit was willing but the body is weak.'"
4. ₿ Crypto Is Borderless. Its Owners Are Not.
Henley & Partners' Crypto Wealth Report 2026 counts 135,694 people holding at least $1 million in digital assets, 92,272 of them in Bitcoin, alongside 290 centi-millionaires and 23 crypto billionaires. The global market sits at $2.6 trillion, $1.6 trillion of it Bitcoin, with Bitcoin trading about 38 percent below its October 2025 peak. Some 742 million people now hold digital assets in some amount, so ownership kept widening through the downturn. The accompanying Henley Crypto Adoption Index ranks 36 countries on more than 900 data points: Singapore takes first for the fourth year running, the UAE jumps from fifth to second on a perfect tax-friendliness score, Hong Kong is third on infrastructure, and the United States is fourth with a perfect 10 for public adoption. Dominic Volek, Group Head of Private Clients, put the thesis plainly: "Crypto may be borderless, but the families who own it are not." Seventy-six jurisdictions have signed the OECD crypto reporting framework, with the first data exchanges due in September 2027. (Read More)
🫖 TEA For Thought: "Singapore first, UAE second, and Hong Kong third. They say web3 speaks Chinese, and it couldn't be more true."
5. ✍️ Two Rules for Writing With an LLM
A new essay on sockpuppet.org argues that readers detect LLM phrasing "in the parts per trillion," so an AI-written paragraph registers as output rather than writing, and it offers two rules for using a model as a copyeditor instead of a ghostwriter. Rule one: you may not use a single word the model suggests, because frontier models are stuck in a register where every line reads like a magazine headline. Rule two: forbid encouragement, since a model that praises your first draft talks you out of the rethinks that carry your voice. What models are genuinely good at is flagging the problems you would otherwise catch only through tedious manual passes: passive voice, buried verbs, repeated turns of phrase, filler like "very" and "actually," and the two or three paragraphs that belong somewhere else. The suggested loop is to have the model spot problems, rewrite the passage yourself, then hand both versions to a second model that has no idea which one you just wrote. (Read More)
🫖 TEA For Thought: "Definitely needed. A good reminder that we can only outsource intelligence, never understanding. Writing is where you learn. Don't give that up."
🛠️ Skill of the Day
The Skill Transfer Map: turns what you already know into the fastest on-ramp into a field you don't, and names the expertise that will quietly mislead you.
You are a learning strategist who specializes in negative transfer: the
places where existing expertise quietly makes someone WORSE at a new field.
I am already strong at: [YOUR CURRENT FIELD, plus 2-3 things you are
specifically good at]
I want to become competent at: [TARGET FIELD OR SKILL]
I can give it [HOURS] per week for [NUMBER] weeks.
Give me three lists.
1. CARRIES OVER. Skills, habits and mental models that transfer directly.
For each one, name the word the target field uses for it, so I can
recognize it when I meet it.
2. MISLEADS ME. Things that feel like they transfer but do not, and the
specific wrong instinct each will hand me. This list matters most, so
make it the longest and be blunt.
3. NO ANALOG. What I have to learn from zero, ranked by how early it
blocks everything else.
Then give me a week-by-week on-ramp that starts with list 3, not list 1.
Finish with one question whose answer would tell me whether I am actually
getting better or just feeling busy.
Ask me anything you need before you answer. Never use an em dash.Paste into ChatGPT, Claude, or your tool of choice. Replace the bracketed parts with the field you have and the field you want.
TEAHEE Moment
Stay sharp, stay informed. See you tomorrow.
If you enjoyed this TEA, follow along on social for more: