2024-12-08 at 1331 mimi import

when youve honed your reasoning you can sorta Feel if an equation is off

that feeling is an action tugging (just like how u feel a word on the tip of ur tongue or how you feel) (i rly like model of . viewing feelings as action tuggings … its not perfect but i like it)

and thats bc your Type 2 reasoning is built using Type 1, and Type 1 is multimodal whose teleology (and thus ontology) is to control Actions

how did humans figure out how to reason ? maybe instead of teaching a model how to reason using gradient descent + RLHF + etc like we teach humans, maybe we can just make it evolve to reason

oh also idk if i wrote this down alr but . the distributed compute of humans is rly rly important . one person’s compute is so so so little but as a team + with using tools n tech, we have so much power

also greatness cannot be planned (referring to the book and related ideas) — i think i alr weote abt how many innovations in aesthetics and science have been driven by Not Optimizing Directly (rederived without remembering the book, ig bc i alr love these textures) — but also i think this is related to the above paragraph where we rly need distributed compute to have diverse perspectives

also related, humans literally need to be taught every, and so with this point along with the prev two paragraphs, idk why we’re trying to create One Big AGI, it just seems stupid and counterproductive and not to mention also computationally wasteful. it seems much more Effective (not to mention also Efficient) to create teams of small models

also i swear salena literally has near-zero reasoning ability, and she can only interpolate on math problems that she’s seen other people solve and she can only interpolate on literary techniques that she’s seen other people do .. so if i can figure out how to teach salena to do ARC tasks, then i can figure out how to win ARC …

why am i always attracted to careers with massive risk and kronecker delta payoff

actually not true i also wanna be a high school music teacher

chollet is trying to measure intelligence as skill learning efficiency, and then he’s trying to use ARC to measure that bc it’s like oh you must learn the skill at test time, but no i think this is a bit dumb,, like iq tests r a skill just like any other

wait why tf is chollet saying to do discrete program search

also i agree w david shapiro in this video https://youtu.be/ZogIWnIMK-E? when he says that this is too mathematical and convergent (as opposed to divergent) to actually measure any intelligence of actual substance