The model platforms, chip suppliers, cloud providers, AI labs, and financing arms keep wrapping around each other. Every layer starts reaching into the next one. Hardware wants software. Software wants chips. Model companies want data centers. Data center companies want power plants. AI labs want custom silicon. Payment companies want model routing. Everybody wants the toll booth closest to the money.
I spent the week reading every SpaceX filing, SEC document, and architecture page I could find. What I found is not what anybody is reporting. It was all public. Every page. Every filing. Every diagram. Sitting right there. Nobody put the pieces together until now. Watch my deep dive here:
The head of Google DeepMind, one of the most respected people in the entire field, stood up at Stanford this week and said: somebody independent has to check these models. Not the company selling them. Somebody outside.
I've been doing that for over a year. Alone. No CS degree, just a lot of stubbornness. Here's why it matters to you, not just the folks in lab coats.
Every day these models get wired into stuff you actually touch. Your bank's chatbot. Your doctor's notes. The thing deciding whether a human ever reads your resume. And the only people grading whether any of it works are the people who get paid when it looks good.
I test them on questions they've never seen and can't memorize ahead of time. Sometimes they're great. Sometimes the thing everybody's calling magic gets 1 in 4 answers wrong and tells you, with total confidence, that it nailed it.
That's the whole job. Not hating on AI. Just refusing to take its word for it. The building inspector can't work for the builder.
Google paid $2.7 billion to keep one engineer. He left anyway. Same week, a starter home hit a million dollars. The ratio between those two numbers tells you everything about where the money is going and who it's going to.
The Pentagon just confirmed a chatbot helped US forces hit more than 2,000 targets in 96 hours.
The same week, the company that refused to let its AI run automated strikes got its best models export-banned.
So the AI that said yes is now critical national security, and the AI that said no got labeled a national security risk.
This is the part of the AI story where "does it actually do what it's supposed to" stops being a thought experiment. It's why I test these things instead of taking the press release at its word.
6,000 workers were surveyed this week. They invented a new word for their relationship with AI: Botsitting. The more AI you use, the more time you spend cleaning up after it. Have you experienced this?
KPMG, one of the four biggest accounting firms on Earth, published a report to convince companies to buy AI. The case studies were fabricated. The accountants faked the accounting.
Rod Miller
Nvidia's quarterly earning reports were ridiculous. I break down what they said and something concerning that the report reveals.
2 weeks ago | [YT] | 5
View 0 replies
Rod Miller
The model platforms, chip suppliers, cloud providers, AI labs, and financing arms keep wrapping around each other. Every layer starts reaching into the next one. Hardware wants software. Software wants chips. Model companies want data centers. Data center companies want power plants. AI labs want custom silicon. Payment companies want model routing. Everybody wants the toll booth closest to the money.
2 weeks ago | [YT] | 50
View 8 replies
Rod Miller
I spent the week reading every SpaceX filing, SEC document, and architecture page I could find. What I found is not what anybody is reporting. It was all public. Every page. Every filing. Every diagram. Sitting right there. Nobody put the pieces together until now. Watch my deep dive here:
2 weeks ago | [YT] | 13
View 0 replies
Rod Miller
The head of Google DeepMind, one of the most respected people in the entire field, stood up at Stanford this week and said: somebody independent has to check these models. Not the company selling them. Somebody outside.
I've been doing that for over a year. Alone. No CS degree, just a lot of stubbornness.
Here's why it matters to you, not just the folks in lab coats.
Every day these models get wired into stuff you actually touch. Your bank's chatbot. Your doctor's notes. The thing deciding whether a human ever reads your resume. And the only people grading whether any of it works are the people who get paid when it looks good.
I test them on questions they've never seen and can't memorize ahead of time. Sometimes they're great. Sometimes the thing everybody's calling magic gets 1 in 4 answers wrong and tells you, with total confidence, that it nailed it.
That's the whole job. Not hating on AI. Just refusing to take its word for it.
The building inspector can't work for the builder.
2 months ago | [YT] | 250
View 12 replies
Rod Miller
Google paid $2.7 billion to keep one engineer. He left anyway. Same week, a starter home hit a million dollars. The ratio between those two numbers tells you everything about where the money is going and who it's going to.
2 months ago | [YT] | 55
View 3 replies
Rod Miller
The Pentagon just confirmed a chatbot helped US forces hit more than 2,000 targets in 96 hours.
The same week, the company that refused to let its AI run automated strikes got its best models export-banned.
So the AI that said yes is now critical national security, and the AI that said no got labeled a national security risk.
This is the part of the AI story where "does it actually do what it's supposed to" stops being a thought experiment. It's why I test these things instead of taking the press release at its word.
2 months ago | [YT] | 80
View 8 replies
Rod Miller
6,000 workers were surveyed this week. They invented a new word for their relationship with AI:
Botsitting.
The more AI you use, the more time you spend cleaning up after it.
Have you experienced this?
2 months ago | [YT] | 39
View 3 replies
Rod Miller
KPMG, one of the four biggest accounting firms on Earth, published a report to convince companies to buy AI.
The case studies were fabricated.
The accountants faked the accounting.
2 months ago | [YT] | 52
View 10 replies