The Diary Of A CEO

AI safety researcher Roman Yampolskiy predicts AGI around 2027 and humanoid robots capable of replacing physical workers by 2030. He warns unemployment could eventually reach 99%, although he distinguishes the technological ability to replace workers from how quickly businesses actually adopt it.

He sees no reliable career to retrain for as AI advances. Programming, prompt engineering, and AI-agent development are all becoming automated, leaving only small markets where customers specifically prefer human service; despite the episode's title, he identifies no five guaranteed surviving jobs.

After 15 years researching AI safety, Yampolskiy now believes permanent control of superintelligence is impossible. He cites unpredictable model behavior, inadequate safety restrictions, and OpenAI's abandoned superalignment team, which had originally announced a four-year research goal.

He argues existing AI could deliver decades of economic growth without creating superintelligence. Yampolskiy estimates current models could already replace 60% of jobs, while 99% of their potential economic value remains unrealized; he wants companies to exploit those capabilities instead of racing toward uncontrollable systems.

Yampolskiy questions Sam Altman's commitment to safety and identifies financial incentives behind departures from OpenAI. Discussing Ilya Sutskever's departure, he cites employees' concerns about Altman and claims former OpenAI researchers can attract startup valuations approaching $20 billion without established products or customers.

He considers AI-assisted biological attacks a plausible route to mass casualties even before superintelligence arrives. Malicious actors could use advanced systems to develop novel viruses, while a genuinely superintelligent system might discover destructive methods beyond human understanding.

Yampolskiy is almost certain humanity lives inside a computer simulation. He argues that future civilizations could run billions of conscious virtual worlds, vastly outnumbering original realities, and personally proposes creating a billion simulations of this very interview once doing so becomes inexpensive.

He owns Bitcoin because he believes AI-driven abundance will make genuine scarcity increasingly valuable. Bitcoin's 21-million-coin supply limit appeals to him more than physical commodities that future technologies might produce cheaply; he even considers investments designed to pay off a million years from now.

Geoffrey Hinton estimates a 10–20% chance that superintelligent AI will wipe out humanity, potentially emerging within 10–20 years. He admits the extinction estimate is intuition rather than scientific calculation and urges governments to force AI companies to invest heavily in safety research.

Hinton believes digital AI has a fundamental advantage over human intelligence: it can share learning directly between copies. Identical networks can exchange enormous amounts of information and preserve their knowledge across hardware replacements, a realization that helped convince him AI could eventually surpass humans.

Hinton believes Ilya Sutskever left OpenAI over safety concerns, although he has no inside information about his departure. They still meet for lunch in Toronto, and Hinton suspects OpenAI reduced resources promised for safety research; investors subsequently committed billions to Sutskever's new AI safety company.

Google did not silence Hinton or force him out over his AI safety warnings. He retired at 75, partly because programming had become more difficult, and timed his departure to speak without corporate self-censorship; Google had offered him freedom to continue safety research.

Bartlett says a major company's CEO privately described plans to reduce its workforce from over 7,000 to 3,000 employees as AI replaced tasks. AI agents reportedly handled 80% of customer-service inquiries, while Hinton's niece reduced complaint-letter processing from 25 minutes to five using a chatbot.

Hinton expects widespread job displacement and recommends plumbing over routine office work as a safer career choice. Physical manipulation remains difficult for robots, while universal basic income could prevent poverty without replacing the purpose and dignity that employment gives people.

Hinton warns that autonomous weapons could make invasions more frequent even when the machines operate exactly as intended. Replacing soldiers with robots reduces the domestic political backlash from casualties, while European AI regulations exclude military applications from their principal protections.

Hinton sold DNN Research to Google at 65 largely to secure the financial future of his son, who has learning difficulties. After losing two wives to cancer, he says his greatest personal regret is spending too much time working instead of with his second wife and children.