From MarketWatch, November 4:
This overlooked area of tech is making money for investors
many times the letter “R” appears in “strawberry.”
OpenAI’s o3 just cleared artificial general intelligence (AGI) benchmarks. Eighty-seven percent on ARC-AGI, the test that’s supposed to measure whether machines can actually think.
Silicon Valley popped the champagne. OpenAI’s Sam Altman took a victory lap. The headlines screamed “AGI is here!”
Except there’s one tiny problem. The same AI that aced the genius test can’t count how many times the letter “R” appears in “strawberry.”
It’s three, by the way. S-T-R-A-W-B-E-R–R-Y. Three R’s. But advanced language models (the ones that can write legal briefs and debug code) insist there are two.
My 4-year-old grandson can count the three Rs in strawberry. I could reward him with Skittles. OpenAI’s o3 costs up to $30,000 per task and still can’t figure it out.
AI fails at character-level reasoning because of how it processes language as “tokens” rather than individual letters. When you can write a sonnet but can’t spell “strawberry,” you haven’t achieved intelligence. You’ve achieved an extremely expensive parlor trick.
It’s like teaching a dog to play poker, then discovering it can’t count the cards. Impressive? Sure. Useful? Not if you’re playing for money.
This isn’t a bug. It’s a metaphor for America’s entire AI strategy. We’re winning a race to build artificial gods that can’t count. China is winning the race to deploy AI that actually works.
But some middle-layer companies are bridging this gap. They’re about to print money. And Wall Street is missing it entirely.
The Wrong Holy Grail
China isn’t trying to build artificial deities. It’s embedding good-enough
AI into manufacturing, logistics, and infrastructure.Former Alphabet CEO Eric Schmidt calls artificial superintelligence “tech’s holy grail.” In congressional testimony this year, Schmidt warned that America’s AI sector would need power equivalent to 90 nuclear plants by 2030.
Meanwhile, a report in Foreign Affairs, entitled “The Cost of the AGI Delusion,” argued that, by chasing superintelligence, America is falling behind. China isn’t trying to build artificial deities. It’s embedding good-enough AI into manufacturing, logistics and infrastructure — with 70% adoption targets by 2027.
Yes, American companies are deploying AI aggressively. Microsoft’s Copilot, Salesforce’s Einstein — every major enterprise is racing to integrate LLMs. U.S. tech companies lead in cutting-edge AI research.But here’s the uncomfortable truth: More than 80% of U.S. AI projects fail to deliver results. Eighty-eight percent of pilots never reach production. The issue isn’t whether America is using AI — it’s whether the U.S. can deploy it at scale. China’s treating AI adoption like high-speed rail: centrally coordinated, massive infrastructure investment and mandatory targets. America is letting each company figure it out independently.
U.S. companies are trying to build superintelligence that can’t count to three. China is building infrastructure that works right now.
As President Donald Trump’s AI czar, David Sacks, admitted: “China is not years and years behind us in AI. Maybe they’re three to six months.” Real impressive lead we’ve got there.
A recent analysis coauthored by Schmidt; Alexandr Wang, now chief AI officer at Meta Platforms; and Dan Hendrycks, director of the nonprofit Center for AI Safety, warned that if either the U.S. or China approaches superintelligence first, the other would view it as a national-security emergency, potentially triggering preventive cyberattacks or even kinetic strikes on AI datacenters.America is betting civilization on a sprint to build something it doesn’t understand while China wins the race to deploy AI that works.
The missing ingredients nobody’s fixing
Three fundamental problems stand between automation and actual reasoning:
1. Autonomous self-learning: Current AI needs massive training runs consuming millions of dollars in compute. It can’t learn continuously like humans do....
....MUCH MORE
