Entry 270-2 Mastery is a System 1 min ↩ back to the timeline

Push back on the machine

The rule that makes the whole AI-as-partner method work, stated plainly: never take what Claude says as true just because it said it. The model is relentlessly positive, so if something doesn't sound right, you stop and push back. Juan calls it 'forcing': Claude tried to typecast him and he said no, this isn't me, and pushed. That same back-and-forth is how he traced the chest pain to costochondritis instead of accepting the first answer, and it's the exact advice he gave Julia for her posts: if what Claude wrote for you to publish doesn't sound like you, don't publish it, push back and say what you didn't like. Good results with AI come from the friction, not from swallowing the first, flattering output.

video fuente → Source video thumbnail
Source transmission · “0 to 1 Million” diary

// trace: where this idea came from

The compass only works because of a rule Juan states as plainly as he states anything: you do not take what Claude says as true just because Claude said it ▸ 6:02. The model is relentlessly positive; it will happily tell you “yes, you are exactly that kind of person,” and you start to believe it, and that is the trap. So the discipline is the opposite of trust: if something doesn’t sound right, you stop, and you push back a little.

He has a word for it, “forcing.” Claude tried to box him into a category during the compass conversation, and he answered, out loud, “hey, no, this isn’t me,” and pushed ▸ 6:26. The pushing is not stubbornness; it is how the real answer surfaces. It is the same back-and-forth that once took him from a frightening chest pain to a name, costochondritis, instead of stopping at whatever the model offered first.

forzar: si no suena como usted, no lo publique →

And it is the exact advice he gave Julia, which is what makes it a method and not just a mood. If what Claude drafts for her to publish doesn’t sound like Julia, doesn’t sound like her, then she shouldn’t publish it: push back, tell it what you didn’t like, this line and that line ▸ 6:37. That, he says, is really how you get good results out of artificial intelligence. Not by taking the first output because it flatters you and calls itself the best. The value is in the friction. The machine proposes; you refuse until it sounds like you; the thing that survives the refusal is yours…

// continued in

la lista es el activo →

Continues

One distilled teaching per entry, straight from the video diary. Leave your email and get the next one the day it lands.