The zebrafish objection

His exhibit is a famous recording: the complete neuronal activity of a zebrafish, every impulse lit in red ▸ 58:08. What it shows is a storm, “miles de cosas al tiempo”, cascades everywhere, whole regions igniting on a stimulus, in an animal carrying an infinitesimal fraction of a human’s neurons ▸ 58:56. Against that, his description of the machines he programs daily: “funcionan de izquierda a derecha”, one token in, one token out ▸ 59:29.

The objection is architectural, not vibes: current AI is unidirectional where a brain is always on, “nunca hay una neurona latente que se prenda aquí en un momento” ▸ 1:00:25, and he grants the standard rebuttal (each pass can feed the next) before rejecting it as a simulation of recurrence, not the thing. His falsifiable stake in the ground: “hasta que se logre entrenar una IA que pueda dar un feedback loop para atrás, no vamos a llegar a una AGI” ▸ 1:01:03.

el pez cebra dispara en todas direcciones; el transformer solo lee →

Which reframes the AGI discourse as what he thinks it currently is: “AGI por el momento termina siendo simplemente un boost”, a fundraising word ▸ 1:01:50, wielded hardest by an unprofitable company that must keep the hype alive. His label sticks: OpenAI is becoming the Apple of AI ▸ 1:02:20, positioned as the disruptor so thoroughly that every minor release detonates, GPT-5’s hyped launch and its infamous charts included, while the pack that ships quietly has already caught up ▸ 1:03:09. Whether his AGI timeline is right, the method holds: judge the claim against the architecture, and judge the announcer against their revenue model…