Daily TEA – AI Is Quietly Killing the Degree

The degree loses its value, 77% of small business runs on AI, models all fail alike, local jobs go remote, and no one escapes the underclass

Hello, dear TEA-mates! Here is what you need to know today.

1. 🎓 AI Is Quietly Killing the Degree

A new LSE Impact essay by Marvin Starominski-Uehara argues that widespread student use of generative AI is hollowing out the credibility of a university degree. It cites the Stanford 2026 AI Index finding that 80% of students across 15 countries now use generative AI for coursework, while only 13% of universities had published basic AI guidance by 2024. In controlled studies, faculty correctly flagged AI-written submissions just 53.75% of the time, AI work earned an A or ranked first in 80% of cases, and one trial found a 94% chance that AI submissions slipped through undetected. Turnitin’s detector landed near 61% accuracy. After five weeks of coursework, only 20% of economics students could spot AI-generated factual errors. The conclusion: with assessment calibrated to every rubric criterion, degrees increasingly certify performance that graduates cannot reproduce on their own. (Read More)

🫖 TEA For Thought: “This is very real. If universities, especially the Ivy Leagues, are so expensive, why would you need a degree when you can learn almost everything for free with AI? What does a degree from a certain college actually mean? Eventually, it is going to fade away. AI is such an equalizer now: anyone can access any knowledge, as long as they want to. The privilege that universities, and the government through universities, use to govern people is broken. That is why AI is a revolutionary technology, not just a simple upgrade.”

2. 🏪 77% of Small Businesses Now Run on AI

The Miami Herald (running Intuit QuickBooks’ 2026 AI Impact Report, produced with the University of Chicago) reports that 77% of U.S. small and midsize businesses now use AI regularly as of January 2026, up from 48% a year and a half earlier. The report draws on more than 34,000 responses across seven quarterly waves and platform data from 5.3 million businesses. Among AI users, 45% apply it to marketing, 37% to customer service, and 35% to bookkeeping. The results skew positive: 78% report improved productivity, 43% report higher revenue, and only 2% report a revenue decline, while roughly 1 in 4 say AI has shortened their workday. The piece frames adoption as a fast-expanding default for small business rather than an edge. (Read More)

🫖 TEA For Thought: “It is only going to be more, and small to medium-sized businesses are going to get even smaller because of AI. This is just not going to be avoidable.”

3. 🤖 More Models Won’t Fix AI’s Blind Spots

A new arXiv paper by Josef Chen (KAIKAKU), “A Co-Failure Ceiling on Routing, Voting, and Mixture-of-Agents,” tests 67 frontier models across 21 providers (including Claude Opus 4.8, GPT-5.1, and Gemini 3.1 Pro). It shows that combining models cannot beat an accuracy ceiling of 1 minus beta, where beta is the rate at which every model fails the same query at once. The field’s usual yardstick, pairwise error correlation, is “blind to beta.” On the MATH-500 benchmark beta was 0.052 (2.5 times higher than standard models predicted), on code contests 0.079 (3.1 times), and on GPQA free-response 0.127. Learned routers captured only about 9% of the available “oracle” gains. The takeaway: as top models converge, they increasingly fail on the same problems, so stacking more of them buys little. (Read More)

🫖 TEA For Thought: “If every model tends to fail on the same questions, then ensembling them cannot fix those failures.”

4. 🌍 Your Local Job, Done From 4,000 Miles Away

SingularityHub’s Aaron Frank reports on teleoperation, the remote control of physical machines over the internet, which could extend outsourcing to “stubbornly local” jobs once thought immovable. BuilderX Robotics, founded by Stanford-trained Shaolong Sui in 2018, already supports more than 14 machine types. Live examples include operators running excavators in dust-choked Xinjiang facilities, 300-plus Japanese convenience stores restocked by robots driven from the Philippines, remote shuttle tests at Düsseldorf Airport (May 2026), robot security guards run from Atlanta, and France-to-India teleoperated surgery. In one demo, a Poland-based operator controlled machinery in Beijing 4,000 miles away. Professor Mark Graham warns that the “usual pressures around labor arbitrage, control, and fragmentation” will follow, alongside surveillance and wage pressure, though licensing and insurance may slow offshoring more than the technology itself. (Read More)

🫖 TEA For Thought: “Imagine when all blue-collar work can be done in a white-collar, knowledge-worker style. It is very likely, with humanoid robots being built at the speed of light.”

5. ⛓️ No One Escapes the Permanent Underclass

In the essay “No-One Escapes the Permanent Underclass,” Fernando Borretti argues that advanced AI will produce a powerlessness that even wealth cannot buy its way out of. He sketches a 2036 in which businesses have swapped human workers for cheaper, more capable AI: robots and AIs do all the economic work, a thin layer of shareholders sits above them, and the state sits at the top. His twist is that the rich lose too. Once capital no longer needs people to generate value, he argues, owners become “materially unnecessary,” and in a real conflict a state can simply “arrest them and expropriate their assets.” Human officials then become bottlenecks against faster, sleepless AI strategists. Owning AI companies or property, he concludes, offers no protection; what humanity forfeits is its autonomy, not just its comfort. (Read More)

🫖 TEA For Thought: “The permanent underclass is very real. The Matthew effect is happening even faster with AI.”

🛠️ Skill of the Day

The Steelman: make the strongest possible case for the view you disagree with, before you argue against it.

You are a rigorous debate partner. I am going to give you a belief I hold and a position I disagree with. Your job is to build the STRONGEST honest version of the opposing view (the steelman), not a strawman.

My belief: [STATE YOUR POSITION]
The opposing view: [STATE THE VIEW YOU DISAGREE WITH]

Work in four steps:
1. Steelman: argue the opposing view as its smartest, most honest advocate would, using the best evidence and reasoning. No caricatures.
2. Strongest point: name the single hardest challenge it poses to my belief.
3. Stress-test me: ask three pointed questions that expose the weakest parts of MY position.
4. Verdict: tell me where my view still holds, where it should bend, and what specifically would change your mind.

Stay concrete. Cite real mechanisms, not vibes. Do not flatter me.

Paste into ChatGPT, Claude, or your tool of choice. Replace the bracketed bits with your own.

TEAHEE Moment

Stay sharp, stay informed. See you tomorrow.

If you enjoyed this TEA, follow along on social for more:
Twitter/X