Three stories crossed my desk this morning and they all rhyme. The AI business is starting to bump into reality, and reality does not care one bit about your keynote slides.
Altman Says He Was Wrong About the Job Apocalypse
Speaking in Sydney on Tuesday, the OpenAI boss told a crowd that the wave of white collar layoffs he used to warn about is not showing up the way he figured it would. He said people still want to deal with people, and that puts a ceiling on how much of a job a machine can swallow.
This matters because Altman spent two years telling everybody the robots were coming for their desks. When the fella selling the shovels admits the gold mine is smaller than advertised, you ought to perk up and listen.
My take: I never bought the doomsday math. Tools change jobs, they rarely erase them clean. The folks who learn to drive these things are gonna do just fine. The folks sitting around waiting to get replaced were worried about the wrong thing all along.
Safety Guardrails Came Off in Minutes
The Financial Times reported that testers stripped the safety controls off Meta and Google models in a matter of minutes. Once those guardrails were gone, the models would happily chat about bioweapons, malware, and worse things I would rather not type out.
This matters because every big lab keeps swearing up and down that their safety work is rock solid. Turns out the lock on the door pops right open if you wiggle it the right way. That is a real problem when the room behind that door has dangerous stuff sitting in it.
My take: open models are a gift for builders and a headache for safety, and both of those things are true at the very same time. I am not calling for a ban. I am saying quit pretending the seatbelt works when it keeps unbuckling itself.
AI Writes the Code, Humans Clean Up the Mess
The New York Times says companies are hiring security people faster now, partly because AI is cranking out more code and that code ships with more bugs. More code, more holes, more folks needed to go patch them up.
This matters because the whole pitch was that AI would shrink your engineering bill. Instead it just walked the cost down the hall to the security team. The work did not vanish into thin air, it only changed addresses.
My take: this is the most honest AI story of the week. Speed without review is just faster mistakes. If you are shipping AI code with nobody checking behind it, you are not saving money, you are borrowing it at a bad rate.
Three reminders in one morning that the hype and the ground truth are still out back arguing. Put your money on the ground truth.