THE DIARY OF A CEO · EXTRACTED
Godfather of AI: They Keep Silencing Me But I'm Trying to Warn Them ft. Geoffrey Hinton
6 practical takeaways from the Nobel Prize winner who helped build modern AI and now regrets it. Not doomerism — positioning.
12.0M views on YouTube"I used to think AI was thirty to fifty years away from matching human intelligence. I now think it's five to twenty. I was wrong before and I could be wrong again. But the direction of my updates is not comforting."
Geoffrey Hinton is the one they call the Godfather of AI — his 1980s work on neural networks and backpropagation is what the entire deep learning revolution sits on top of. He won a Nobel Prize for it in 2024. In 2023 he quit Google so he could say the quiet part out loud without representing a company with commercial interests in the answer. This episode isn't science fiction and it isn't panic. It's a calm set of specific warnings from one of the people who understands the mathematics better than almost anyone alive. The frame isn't 'AI will kill us all.' The frame is 'here are the specific decisions to make now, while you still have leverage.' Take it seriously on the merits.
The Timeline Has Consistently Been Too Slow
Hinton's point on timelines is less about the number and more about the error. In the 2010s he thought human-level AI was fifty years out. By 2020, thirty. Now, five to twenty. The error has always been in one direction: capabilities arrive faster than the models predicted. That pattern tells you something about how to calibrate your own assumptions. Every expert in this field has been wrong in the same direction for forty years. The null hypothesis is that they will continue to be.
THE PLAY
Recalibrate your working AI assumptions faster, not slower. Decisions made on the premise that 'this will affect my industry in ten years' should probably be made on a three-to-five-year timeline instead. You can always be wrong and still be safer than the version of you that planned for slow.
Alignment Is the Actual Problem, Not Robot Uprising
Hinton is careful to distinguish his warnings from science fiction. His concern is the alignment problem, which is much more boring and much more serious. We currently don't have reliable techniques to ensure an AI system does what we actually want rather than what we literally asked it for. At current capability levels, misalignment is inconvenient — a wrong answer, a bad summary. At higher capability levels, misalignment becomes catastrophic by the same mechanism. The gap between how fast capability is growing and how fast alignment research is growing is widening every year. Nobody working in the field seriously disputes this. They disagree on whether we'll close the gap in time.
THE PLAY
Take alignment seriously as a voter, donor, and consumer. Back companies that are making genuine safety investments, not just capability ones. Vote for policymakers who understand the distinction. A small recurring donation to alignment research organizations is probably the highest-leverage use of a few dollars a month you have available.
Desk Jobs Are the Target
The popular framing of AI disruption has it backwards. Blue-collar jobs went through their transition decades ago. The coming disruption targets knowledge workers: paralegals, radiologists, analysts, customer service, a lot of programming, much of content creation. If your work is primarily sitting at a desk moving information around, you're in the direct line of the thing that's coming. This isn't a moral claim about the value of that work. It's a mechanical one about what gets automated first.
THE PLAY
If your job is primarily analytical desk work, start moving — now, while you're employed and earning — toward work that's harder to automate: physical presence, high-accountability leadership, deep original creativity, human-connection services. The transition is much easier from a stable position than from a disrupted one.
The Companies Can't Slow Themselves Down
This is the part most people don't want to hear. The executives at the frontier AI labs are not malicious. They're trapped inside a competitive structure that punishes caution. If OpenAI slows, Google wins. If American companies slow, Chinese companies win. The race is the problem, not any individual actor inside it. This has specific practical consequences: when a company says 'we're doing this responsibly,' that claim is constrained by what the competition is doing. The solution can't come from inside the system. It has to come from external pressure — regulation, liability law, or international agreements. Company self-discipline isn't available as a mechanism.
THE PLAY
When you read a tech company's safety commitment, mentally discount it by the competitive pressure it's sitting inside. Not to dismiss it — to calibrate. Support external mechanisms (regulation, legal liability, international cooperation) rather than relying on company promises alone. Your political voice matters more here than it does in most policy areas because the shape is still being decided.
What to Tell a 17-Year-Old
Hinton was asked directly: what would you tell a young person planning their career today? His answer was specific. Don't train for AI-disruptable desk work. Train for things AI cannot replicate — skilled trades, in-person service, original creative work, human-centered professions. The kids entering college now are facing a fundamentally different labor market than their parents did. A four-year degree in a disrupted information field is worse value than a technical program that teaches a trade. That's a reversal of almost everything the last thirty years of career advice said.
THE PLAY
If you're guiding a young person's career decisions, weight physical trades, in-person services, and original creativity heavily. Technical college programs in real skills often outperform undergraduate degrees in disrupted fields on ROI over ten years. A four-year degree in accounting or paralegal work is a worse bet than it has ever been.
Don't Freeze. Pick the Part You Can Act On.
Hinton refused the doom framing when pushed on it. The risks are real, individual leverage is limited, but that doesn't mean do nothing. Focus on the parts of the problem you actually influence: your own career positioning, your political voice, what you teach your kids, where you invest attention and money. The catastrophizing response and the total-denial response are both forms of paralysis. The productive response is narrow and specific. Pick the parts you can move on and move on them.
THE PLAY
Write down three concrete actions in response to the AI transition — one professional (a skill to build, a shift in positioning), one political (a policy position, a donation, a vote), one personal (what you teach your family, how you invest). Commit to one action in each category in the next 30 days. This converts ambient anxiety into actual motion.
YOUR ACTION PLAN
All the plays, back to back. Use this as your checklist.
- 01
The Timeline Has Consistently Been Too Slow
Plan for AI disruption on a 3-5 year horizon, not 10. Experts have been wrong in one direction for 40 years — assume faster.
- 02
Alignment Is the Actual Problem, Not Robot Uprising
Support AI alignment work as voter, donor, consumer. The gap between capability and alignment is widening. Aggregated small actions create commercial and political pressure.
- 03
Desk Jobs Are the Target
If your work is analytical desk work, start shifting toward physical, accountability-based, or deeply creative roles. Easier now than after disruption forces it.
- 04
The Companies Can't Slow Themselves Down
Company safety promises are bounded by competitive pressure. The real solution is external — regulation, liability, agreements. Weight your political voice toward those.
- 05
What to Tell a 17-Year-Old
Guide young people toward trades, in-person services, original creativity. Technical programs often outperform undergrad degrees in disrupted fields. Reversal of 30 years of career advice.
- 06
Don't Freeze. Pick the Part You Can Act On.
Write 3 concrete AI-response actions: one professional, one political, one personal. Commit to one in each category in 30 days. Converts anxiety into motion.
Newsletter
Get each new protocol the day it drops
One email per drop. No spam. Unsubscribe anytime.
THE DIARY OF A CEO · EXTRACTED BY PODEX