RT @mustafasuleyman: We're taking a big step towards medical superintelligence. AI models have aced multiple choice medical exams – but rea….

Just remembered that Fraggles literally snacked on the Doozers’ entire housing market like popcorn.

Nothing says ‘trust AI’ quite like Qantas having an entire movie category called ‘A.I. DOOM’ on their flights.

@tomstello_ @GordPennycook @DG_Rand @BrooklynCorbett cc: @karpathy @alexalbert__ @AnthropicAI @drew_bent @AngieMcData @salkhanacademy @emollick @dwarkesh_sp @_sholtodouglas @AmandaAskell @TrentonBricken @demishassabis @rose_e_wang @OfficialLoganK @FryRsquared @patpat_mit @andy_matuschak @sama.

@tomstello_ @GordPennycook @DG_Rand @BrooklynCorbett Pre-registration and all our materials, data, and analyses are open on the OSF: [.

@tomstello_ @GordPennycook @DG_Rand The technology exists. The evidence is here. What we do with it—that’s the only question left. Full paper: [.Authors: @BrooklynCorbett @tangenjm.

@tomstello_ @GordPennycook @DG_Rand For those building the future of education: conversational AI does what no textbook can—it listens, adapts, and argues back. It scales the Socratic method. It makes every learner defend their assumptions, not just consume corrections.

@tomstello_ @GordPennycook @DG_Rand But “learning styles” and “10% brain use” don’t share a worldview. They’re independent errors, each requiring its own conversation. No cascading enlightenment. Just patient, targeted work.

@tomstello_ @GordPennycook @DG_Rand Here’s what surprised us: fixing one myth didn’t fix others. When Costello’s team debunked conspiracy theories, belief in unrelated conspiracies also dropped—a spillover effect.

@tomstello_ @GordPennycook @DG_Rand But defending your belief to a persistent, polite AI? That forces you to articulate why you believe what you believe. And sometimes, hearing yourself explain it is enough.

@tomstello_ @GordPennycook @DG_Rand What’s happening cognitively? Myths persist because they feel right—they emerge from mental shortcuts we all use. Reading facts doesn’t disrupt this.

@tomstello_ @GordPennycook @DG_Rand Like a new exercise routine, the initial burst of change needs reinforcement. One conversation starts the process; sustained change requires more.

@tomstello_ @GordPennycook @DG_Rand Two months later, we checked back. Claude’s advantage had faded—both AI and textbook readers maintained their improved beliefs, but the gap between them had narrowed considerably.

@tomstello_ @GordPennycook @DG_Rand The conversation held their attention in a way textbook passages couldn’t. When someone’s actively defending their beliefs, they can’t skim or tune out.

@tomstello_ @GordPennycook @DG_Rand The numbers tell one story. The transcripts tell another. When someone insisted “different people learn differently,” Claude didn’t lecture. It asked: “What convinced you of that?” Then it gently untangled their reasoning, using their own examples against them.

@tomstello_ @GordPennycook @DG_Rand The personalised AI reduced false beliefs significantly more than textbook reading, which itself outperformed our control.

@tomstello_ @GordPennycook @DG_Rand So we pitted three approaches against each other:.- An AI that engaged people in personalised debate.- A textbook excerpt with all the right facts .- An AI chatting about cats (our control). 375 people. Real beliefs. Real conversations.

@tomstello_ @GordPennycook @DG_Rand Traditional debunking fails because myths don’t live in the fact-checking part of our brains. They nest in our intuitions, wrapped in anecdotes and common sense. *Full item wording in the paper.

@tomstello_ @GordPennycook @DG_Rand Consider the “10% brain” myth. You’ve seen the movies; maybe you’ve corrected someone at a party. Yet our participants—educated adults—gave it an average truth rating of 46/100. Strong enough to influence decisions, weak enough to avoid scrutiny.

Building on @tomstello_ @GordPennycook @DG_Rand’s breakthrough Science paper on conspiracy beliefs, we wondered: could the same approach work on everyday academic myths?.