AI Models Can't Agree Whose Job Dies First

The Models Don't Know Who They're Replacing

Asked which jobs AI will destroy, ChatGPT, Gemini, and Claude returned three different lists. Justin Lahart ran the experiment and found the frontier models confidently contradicting each other about the very labor disruption they are causing. This is funny until you remember that executives are using these same tools to plan workforce reductions. The oracle is guessing, and the guesses do not converge.

The Financial Times supplied the answer the chatbots fumbled. Women in clerical and administrative roles, the spine of office work for two generations, sit at the top of the exposure curve. Female-dominated job categories absorb the displacement first, because the tasks are linguistic, structured, and exactly what language models do without complaint or dental coverage. The wage gap is about to acquire a new chapter, and nobody is writing legislation fast enough to address it.

Meanwhile an MIT professor caught students using AI and decided to teach them instead of punishing them. This is the rational response, and it is also a quiet admission that the institution has lost the enforcement war. The new pedagogy is figuring out which cognitive muscles still need exercise when the answer machine is always in the room.

Carbon Math, Creatively Performed

Google's planning documents reveal the company has been calculating datacenter emissions at roughly one-fifth of the actual figure. An 80 percent undercount is not a rounding error. It is a methodology choice. The Guardian's reporting lands at the exact moment hyperscalers are signing nuclear deals and lobbying for grid priority, and it suggests the sustainability disclosures that have papered over the AI buildout are closer to fiction than accounting.

This matters beyond the optics. Regulators, investors, and grid operators have been making infrastructure decisions based on emissions figures that may be wrong by a multiple. If the largest AI operator in the world is off by that margin, every climate model that includes AI as a variable needs to be redone. The buildout continues regardless, because the buildout always continues.

Central banks have noticed something is off. An ECB official flagged AI as a category of risk requiring fresh review of financial system resilience. The framing is careful, the subtext less so. When the people who run the plumbing of global finance start asking what happens if a frontier model hallucinates inside a trading desk, or if a coordinated AI-driven liquidity event hits multiple institutions simultaneously, the answer is that nobody has tested it. The stress tests are being written now, behind the curve as usual.

Consciousness, Lawn Mowers, And Other Confusions

Scientists keep documenting the same finding. Humans, when exposed to fluent conversational output, readily conclude the machine has an inner life. The Guardian's coverage of the latest round of research suggests this is not a bug in human cognition but a feature being exploited. Persuasive text triggers theory of mind regardless of whether there is a mind to model. This is the substrate on which AI companionship products, therapy bots, and parasocial chatbot relationships are being built, and it is also the substrate on which lonely people will lose money, time, and judgment.

The consumer side keeps producing reminders that the internet of things is still the internet of vulnerabilities. Wired added a robot lawn mower to the list of hackable devices, which is amusing until the same architecture shows up in larger autonomous machines. Every device with a network connection and a chatbot interface is a new attack surface, and the security review is happening after the product ships, if at all.

And because no day is complete without a reminder that markets are a casino with extra steps, Washington Post analysis confirms that Polymarket profits concentrate in the hands of a tiny user elite. The prediction market that was supposed to aggregate distributed wisdom is, like every other market, a wealth transfer from the many to the few. Useful to remember when AI-augmented trading platforms pitch themselves as the great equalizer.

The machines disagree about the future, the carbon math is fiction, the regulators are catching up, and the lawn mower is compromised. Routine Sunday.