rime @rime@schelling.pt

Are the smartest systems today (brains/AI) near the limit of how much a world-model can be compressed, or are we hitting the limit? Given all the data we observe, how many intrinsic dimensions is it distributed over?

@cosmiccitizen otoh, if—more realistically—my rationalization-strength wrt unendorsed motivations was merely extremely high, i wud focus on trying to understand and gain leverage over them, instead of blindly "compensating" wo understanding.

a balancing strategy like "ok, spend 30 sec thinking abt pros, and 30 sec thinking abt cons" is blind in the sense that it has no model of the enemy, which also means that it fails to generate opportunities to *learn* abt the enemy.

@cosmiccitizen oh, nice question...! I guess it depends on my expectation over what my unendorsed subconscious motivations are. if at all I have endorsed motivations which conflict w what actions my infinite-rationalization-capability bias me toward, am I just provably ineffectual/non-causal/epiphenomenal?

ju say "a 1000-bit exact specification of X," and I ask "relative to which interpreter?"

I've been misled by naive information theory for a long time. there is no representation-of-thing which "has the structure" of the thing-in-itself. information is only ever the delta btn contexts. communication is always a bridge over inferential distance.

there is vars that more or less dereference as intended, but it's Gricean all the way down.

(thoughts prob counterfactually inspired by niplav)

OMG

the Nature function herself takes arbitrary coordinates in spacetime, and returns exact distributions of matter within those bounds.

"Nature speaks differential equations" pfff!! differential equations are like gobbledygook to her, and she takes offense at the notion that this is *what she's made of*.

> "Since Newton, mankind has come to realize that the laws of physics are always expressed in the language of differential equations."

this is wrong. differential equations is what we must *settle for* when finding the generators of the data is intractable. if u knew Nature herself, u wudn't be restricted to computing her step-by-step w infinitesimal step-sizes, u cud j interpolate btn arbitrary points w no loss in accuracy.

- the highest-avg-IQ academic subjects are mathematics and philosophy *because* they're also *less* financially profitable (thus, ppl go into them bc they're rly intellectually interested in them). the statistics doesn't seem to bear this out, but that's bc there are confounders—the underlying pattern still holds. :p

- more idr

- if u concentrate hard on finding ideas related to X, u increase the rate at which u become aware of X-related ideas, but u also decrease the threshold of X-relatedness required for becoming aware of them. thus, if u want to maximize the quality/purity of ur X-related ideas, u may wish to *avoid* looking for them too hard. this is the advantage of serendipity as an explicit search-strategy.

EXAMPLES:

- if u pay ppl to do X, u increase the number of ppl who do X, but u also dilute the field bc now ppl do X for monetary incentives PLUS intrinsic incentives, wheareas bfr it was only the latter.

in the context of AI alignment, it implies that the most aligned training-regimes are also likely to be really weak (and therefore expensive).

the paradox of selection: the purity of your selection-pressure is inversely proportional to its strength.

sorta obvious when phrased that way, but the point is that you can increase your selection-pressure over the true target by additionally selecting for increasingly less precise proxies. this increases the rate of true-positives and decreases false-negatives, but it also increases false-positives (confounders), and sometimes at a higher marginal rate.

tbc, I haven't processed this at all. I just needed to make a salient note to myself bc I'm in a hurry, and don't want to forget my uncertainty abt this. huge-if-true but still v unlikely.

reminder to self: resolve whether actually all of so-called "randomness" or "many-worldedness" in physics is a result of "incompressible logic". I'm torn btn finding this lecture completely baffling or utterly mundane. mby it's utterly mundane but its implications are huge? prob neither. noting this down bc I didn't expect myself to become uncertain abt many-worlds-interpretation based on a *wishy-washy lecture on logic*. WHAT THE HECK.

https://www.youtube.com/watch?v=owHuDGMLk5s&list=PL86ECDEDE3FA8D8D1&index=6

Please consider funding Johannes for alignment research or help make that happen: https://www.lesswrong.com/posts/2D74Ctr5Aj3Sb5f69/fund-me-please-i-work-so-hard-that-my-feet-start-bleeding

I endorse his approach as much as I can endorse anyone. (See e.g. Emrik's comment, and TsviBT's endorsement.)

video clip of me breathing with my eyes.
https://www.youtube.com/watch?v=B_SAb2_rVCs

@niplav o i forgor: "alst"↦"at least".

forgor↦forgor

@niplav I wud *prefer* u did all and posted w u's own name :p

tho I not net-disprefer name-on-post.

I is financially secure and do a fairly indep agenda (q ib to remain indep for alst ~2 add years?), so marginal reputation points is not much usefwl for me.

fm my perspective, utility of q post, is entirely its altruism—q ib is substantial (tho risky), so I hope u publish.

I can prob review/comment/take-questions or smth, tho, if wish.

ib↦"I believe"
q↦anaphor (incl "which")
indep↦〃-endent

@niplav Oh, and I should add just in case: If you plan on writing about it, you're not "scooping" me or any such nonsense. The draft is tiny, 2 years old and I don't plan on picking it up. Yet I really hope somebody makes a good post for the whole ideabag. Please do!