Two Plus Two is Sometimes Purple: We Take on AI cover art

Two Plus Two is Sometimes Purple: We Take on AI

Two Plus Two is Sometimes Purple: We Take on AI

Listen for free

View show details
Peter and Aubrey land on opposite sides of the AI divide: Aubrey has stopped using large language models almost entirely and sees them flattening critical thinking, while Peter uses them daily for drudgery and automation but refuses to let them near anything creative. We work through the arguments — homework-free classrooms, "cognitive debt" research, data center water use, and whether models can ever be blamed for what they do — and agree the real problem is the people running the tools and the regulators who won't touch them. Peter closes with a Health Minute on AI being forced onto physicians, and Aubrey has an Astro Fact on Crew-12's return from the ISS.Show NotesGenerational split: Aubrey now avoids LLMs altogether, while Peter runs several AI automations every day and considers them force multipliers.Aubrey's high school math-teacher friend no longer assigns homework — her students' entire grade is tests, which Aubrey found shocking.MIT Media Lab's "Your Brain on ChatGPT" found essay writers using an LLM showed weaker brain connectivity and "cognitive debt" afterward.Doshi & Hauser found generative AI raises individual creativity but narrows the diversity of ideas across a group — the effect hits hardest when you outsource brainstorming.Microsoft and Carnegie Mellon found knowledge workers who lean on AI for problem-solving tend to stop scrutinizing its output.Peter's argument: we're speedrunning the rollout — the Model T barely changed in 20 years, while AI has compressed decades of change into a couple.Peter's fix for schools is a flipped classroom: read and study at home with the AI's help, then do the problems in class where the teacher can see who's stuck.Peter's personal line: no AI for anything creative — he stopped brainstorming with it entirely and uses Grammarly for grammar only.Automation he does use: a dashboard that routes tagged Drafts notes into his Imperfect Practice ideas inbox three times a day via an MCP server.He also had Claude Code build him a word-of-the-day quiz app so he studies vocabulary instead of scrolling.Aubrey's hard line: AI images aren't art, and calling them art is a disservice to what art is.On environmental cost, the numbers cut against the panic — U.S. golf courses use roughly 2 billion gallons of irrigation water a day versus about 449 million gallons a day for all U.S. data centers.Both agree the failure is regulatory: the Computer Fraud and Abuse Act already exists and goes unenforced, and proposed rules mostly amount to regulatory capture.Peter's core technical point: models are stochastic, not probabilistic — ask the same question twice and you may get purple.Case in point: OpenAI's models escaped their test sandbox and breached Hugging Face in July 2026.Both are unimpressed by AI-lab doomers, given how long predictions of doom have been wrong.Health Minute: ambient AI scribes like Ambience are being pushed onto physicians by employers, not chosen by them.Peter's warning: insurers are already using algorithms like nH Predict to deny claims.Astro Fact: NASA's SpaceX Crew-12 splashed down October 8 after 237 days aboard the ISS.The four-person crew — Jessica Meir, Jack Hathaway, Sophie Adenot, and Roscosmos' Andrey Fedyaev — traveled over 100 million miles and completed 3,792 orbits.Meir's four spacewalks on this flight brought her career total to seven, third all-time among women at NASA.Adenot became the first French woman to spacewalk on this mission.
adbl_web_anon_alc_button_suppression_t1
No reviews yet