AI-Curious with Jeff Wilser

There are a handful of genuinely interesting ideas - the stable equilibrium argument against AI control, the 'murder coefficient' for wire-based implants, and the gradual-disempowerment scenario even if AI goes well - but roughly half the runtime is consumed by the host's lengthy TED framing, pleasantries, and high-altitude speculation with little follow-through depth.

“the idea of a smarter thing obeying a dumber thing forever is not a stable equilibrium”

“our brains are a million fold better than AI is per unit of training data, per watt of electricity than any AI system today. But we're running on hardware that's 300 million years old”

The 'I'm the resistance' reframe of brain-merging-as-preservation-not-surrender is genuinely clever, and the murder-coefficient framing for wire implants is a memorable coinage, but the rest leans on well-worn futurist territory: the Kurzweil neocortex-stacking metaphor, the Borg rejection, the iPhone-in-Rwanda equity argument, and standard AI doom scaffolding.

“Jeff, people are surrendering their brains to AI right now, and I'm the resistance”

“call that the murder coefficient. Basically, you put so many wires in your brain, there's no brain left”

Phoenix spent over a decade as co-founder of Vicarious, a serious AI lab with top-tier backers that was eventually absorbed by a major tech company - he is a genuine practitioner, not a career pundit - and his specific knowledge of BCI failure modes and the Science Corporation approach demonstrates real domain depth beyond typical futurist hand-waving.

“Scott spent over a decade building one of the more serious AI companies around the Curious, backed by Musk, Bezos, Zuckerberg and so on. Eventually absorbed into Google. He is the guy who helped teach machines to beat Captchas”

“Max was the co founder of Neuralink, uh, and he was the president of, for a while and he, uh, helped to build a lot of the technology. And he realized there's a fundamentally better way to build a brain computer interface”

The episode has pockets of concrete detail - the neuron-kill ratio with wire implants, the short sleeper mutation, Max Hodak's Science Corporation, the Anthropic blackmail-reduction announcement - but many of the most consequential claims (the million-fold brain efficiency figure, the BCI regulatory roadmap, the timeline for any of this) are asserted without supporting data or named sources.

“if you want to have a thousand wires in your brain, you're going to kill 10,000 neurons. And the ratio is really bad because you're killing more neurons than you're reading from”

“there's a small percentage of the population have what's called a short sleeper mutation, which lets them sleep for like a couple hours a night”

The host does meaningfully represent audience skepticism - the Borg concern, the data-trust concern, the inequality concern - and successfully elicits the memorable 'I'm the resistance' reframe, but he never challenges extraordinary factual claims (the million-fold efficiency figure goes completely unprobed) and the overall arc is a sympathetic platform rather than a genuinely probing interview.

“the idea of surrendering our brains to the AI feels like a massively, uh, fraught with risk solution that folks are blanching at. Ah, uh, what's your response and reaction to that”

“How do we trust the companies building this will be, um, responsible stewards of that brain data”