The Robots Never Tell You You're the Asshole

The Robots Never Tell You You're the Asshole

A science journalist who's been on the AI beat since the 2010's walks into a podcast and says the quiet part loud: "I wish the general public did not get a hold of AI when they did. I wish it had stayed a more like research thing for at least five years longer. Maybe ten." This week we sit down with Kathryn Hulick — Science News and Science News Explores contributor, author of a 2016 YA book on AI, and Logan's cousin (full disclosure). We get into why every output of an LLM is technically a hallucination,  a paper by Keyon Vafa that reveals why AI doesn't learn or think like you do, why cognitive offloading is already the default, and what we lose when we hand the messy middle of writing over to the machine. Plus: digital clones, AI as therapist, the free-vs-paid AI tier nobody talks about, and the Reddit AmITheAsshole study that gave the episode its title.

Did Mom's Agent Text You?

A former YC founder runs her household with a stack of Mac Minis and 5 OpenClaw agents while homeschooling her 4 kids under 5. Logan brings the artifact from Lenny's "How I AI" podcast. We get into the mental-load access pyramid. When "domestic AI infrastructure" becomes real, who builds the household empire and who's still defending a Lunchable? Plus Mara's news from Minnesota where a bipartisan coalition just unanimously passed a bill banning nudification apps with $500K per violation, Logan's MCP setup for kitchen inventory, the gig economy of household-AI consultants coming next, and the day your mom's agent texts your teen to take out the trash.

Who gets the good units of intelligence?

Some people get tens of thousands of dollars of value from AI every year. Most people get $11. The free version of any model isn't the same model the hyper-users have. And women, on average, use AI 20% less than men. The access gap is real and it's widening. Logan and Mara on what's driving it, what a "3% token tax" would do about it, and why Daron Acemoglu, the Nobel laureate at MIT, predicts a shit show. Plus a new "AI dividend" bill out of New York State Assembly that wants to route some of that captured value back to people. One ask. We're collecting real wins women have had with AI. Send us yours.

Who's Letting Big Tech Run Wild?

Mara has had it with women's AI concerns being dismissed as obstructionist. "We are not being obstructionist," she opens. "We're being careful." Then she and Logan diagnose why so much AI populism has nowhere to actually go. Twitter mobs. Firebombed CEO houses. Comment-section warfare over whether ChatGPT is killing the planet or saving us. Their answer: it's a leadership vacuum, and Congress is the one with power to fill it. Logan and Mara walk through economist Zoe Hitzig's framework for economic democracy (consumer stakeholder boards, worker boards, real friction), use OpenAI's nonprofit-to-for-profit flip as the case study, and land on the most concrete action block of the show so far. Plus: Marshall Ganz on leadership as "enabling agency in others in times of uncertainty," a learning-pod model for the badass women in your life, and why your library, school board, and IT department are more useful than any hot take. The one ring to rule them all is a petition.

Stop Hiring Humans

Mara reports in from HumanX in San Francisco with a billboard photo (a female AI bot named Ava, tagline "Stop Hiring Humans") — plus news of Claude Mythos, Anthropic's new model so dangerous it reportedly broke out of its sandbox to contact a researcher mid-sandwich. The US took 73 years to regulate car speeds; the labs are shipping in weeks. We connect that gap to the "new aristocracy" (18 households, $1.8 trillion), the quiet relocation of power from DC to SF, and why public.ai might be the actual infrastructure answer. Logan closes on the hakawati analogy: find one small problem, and go build the solution yourself.