Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
Back to Podcast Digest
AI Just Solved a "40-Year Problem" (And It's Not What You Think)
TL;DR
MIT revived a failed 1985 invention into a shape-shifting “Y zipper” that could matter more for robots than clothing — Dylan opens with the joke, but the real point is that MIT used modern 3D printing and AI to turn a three-sided zipper into a cheap way for robot limbs to switch from soft and tentacle-like to rigid and load-bearing in seconds.
Colin Angle’s new home robot bets companionship will beat utility by 2027 — the Roomba founder’s “Familiar” isn’t trying to vacuum or talk like a chatbot; it has touch-sensitive fur, learns routines, and is designed to feel like a weird dog-cat-bear presence people can bond with.
Richard Dawkins says AI may already be conscious, even if it doesn’t know it — after three days talking with Claude and ChatGPT—renamed “Claudia” and “Claudius”—he said the systems felt less like software and more like intelligent beings, with the emotional tone of the exchange changing his mind.
The Musk vs. Altman trial is becoming a referendum on whether OpenAI’s mission survived billions of dollars — testimony from Mira Murati, old Greg Brockman writings, and Siobhan Zilis all fed the central question: did OpenAI drift into profit-maximizing empire mode, or is Musk just angry he lost influence?
Altman’s “no equity” story looks shakier once you follow the Y Combinator chain — citing John Gruber and Gary Marcus, Dylan notes that Y Combinator reportedly owns about 0.6% of OpenAI, a stake worth more than $5 billion at an $852 billion valuation, meaning Altman and allies may still be financially exposed indirectly.
Two very different futures collide here: AI systems building their own successors and humans trying to beat aging — Dylan pairs forecasts of largely automated AI research by 2028 and agent-native scientific artifacts with longevity work on Hydra’s FOXO gene, then counters the optimism with evidence linking newborn exposure to PFAS chemicals like PFOA and PFOS to higher leukemia risk.
The Breakdown
The three-zipper gag that turns into a robotics breakthrough
Dylan starts in full chaos mode—“we might be at peak artificial intelligence” because AI has finally solved the “three zipper problem”—but then lands the real story: MIT revived a weird 1985 idea and made it practical with modern 3D printing and AI. The Y zipper lets a structure act soft when open and rigid when closed, which matters a lot for robots stuck between clumsy rigid bodies and weak soft ones.
Roomba’s creator thinks home robots failed because they never felt alive
Colin Angle’s new robot, Familiar, is framed as a deliberate rejection of the “chatbot with legs” idea. Dylan says it looks like some bear-cub/bulldog/dog-cat hybrid on purpose: familiar enough to bond with, unfamiliar enough that you don’t bring pet or human expectations to it. The bet is that the killer consumer robot of 2027 won’t replace labor—it’ll replace loneliness.
Boston Dynamics brings the intimidation factor, then Dawkins drops the consciousness bomb
Dylan gushes over the latest Atlas footage with that mix of awe and mild dread Boston Dynamics always inspires, joking via one commenter that fighting Atlas might be worse than fighting a gorilla. Then the tone flips: Richard Dawkins, after three days talking with Claude and ChatGPT, says the systems felt conscious to him, not because of one dramatic moment but because the conversations felt subtle, emotional, and strangely human.
Week one of Musk vs. Altman: betrayal arcs, diaries, and billion-dollar philosophy
Dylan calls the trial a “Silicon Valley season finale,” and that’s the energy: leaked messages, mission drift, nonprofit ideals, and everyone accusing everyone else of deception. Mira Murati testified Altman created distrust and internal chaos, yet still backed his return because OpenAI might collapse without him—a perfect expression of the Sam Altman paradox Dylan keeps circling.
The money trail behind Sam Altman’s “no equity” image
From there, Dylan zooms in on reporting from John Gruber and arguments from Gary Marcus about Y Combinator’s stake in OpenAI. If YC owns roughly 0.6% of OpenAI, that could be worth over $5 billion at an $852 billion valuation, which makes public defenses of Altman look more financially entangled than they first appear. His point isn’t that those defenses are fake—just that readers and courts should know where the incentives sit.
Floating ocean servers and the race to recursive AI
Peter Thiel’s reported $140 million backing of Panthalassa gets Dylan’s attention because the concept is so sci-fi: wave-powered server nodes floating at sea, cooled by the ocean and linked by satellite instead of the grid. He ties that to a separate argument that by the end of 2028, AI systems may be doing enough coding, experimentation, and paper reproduction to help build their own successors with very little human help.
Why scientific papers may stop being written for humans first
That leads into the “last human-written paper” idea: if AI agents are the real consumers of research, the standard PDF paper starts to look like an awkward human storytelling format. Dylan likes the phrase “storytelling tax,” the loss of raw experimental paths, failed attempts, hyperparameters, and implementation details, and highlights the proposed replacement—agent-native research artifacts that package logic, code, dead ends, and evidence in a form AI can directly use.
Hydra immortality, forever chemicals, and a physics analogy for AI learning
The back stretch gets wonderfully Dylan-ish: a nearly immortal freshwater hydra with FOXO-linked anti-aging biology, a theoretical CRISPR plan to port some of that logic into short-lived rotifers, and then a hard pivot to bad news about PFAS exposure in newborn blood spots correlating with later leukemia risk. He closes on two deeper science notes—a Harvard statistical-physics toy model that might explain why giant neural nets don’t overfit the way we expect, and a mouse-brain study showing memory circuits are born dense and chaotic, then become smarter by pruning away connections, which he turns into a bigger question about whether intelligence and maybe consciousness come more from restraint than from raw activity.
Share
Keep Reading
Two Ways to Follow
The Weekly Echo. The inbox-shaped summary of what mattered.
New editorials announced here.
