The reward function

The splinter left by the failed interview: Girdley said, repeatedly and happily, “me gusta la plata, me gusta hacer plata” ▸ 14:47. Which collides with Julia’s law-of-attraction reading of the week and produces the uncomfortable question: if you attract what you desire, is his problem that he stopped desiring money ▸ 15:05?

The history says he used to. Every job was chosen for pay, he says so plainly, the culture speech is what people tell themselves until a double offer arrives ▸ 17:19. Except the last one: Parameta raised his salary, offered more to stay, and he left anyway, because “la IA me estaba llamando” and the one thing they couldn’t give him was time to study it ▸ 18:53. In that moment: “yo no estoy buscando plata” ▸ 19:43.

el dinero mide el valor presente; lo que estás incubando no aparece en la métrica →

His cousin’s line, once dismissed as reductionist, gets its rehabilitation: money is the simplest comparable metric of the value you generate ▸ 20:41, and for an entrepreneur staring at a horizon with no grades and no promotions, simple metrics are oxygen ▸ 21:39. But he keeps his one objection, and it’s the professional one, straight out of his research: money is a present-value signal that can’t see potential, the studying child earns nothing and is not worthless ▸ 23:26. In reinforcement learning the failure has a name: the short-sighted agent picks actions with good immediate reward and never reaches the long-term optimum, and designing reward functions that don’t do this is an entire, unsolved study ▸ 25:52. His own year was that bet, walking from salary into zero income to acquire a capability ▸ 24:59. Julia’s counter-metric is goals set and reached ▸ 22:27; his synthesis is a reward function with a longer horizon: chase money now, “ya es hora de buscarla” ▸ 20:07, but never let the metric convince you the incubating thing is worthless…