This might be a bit of a hard­ball ques­tion for an event post, but I’ve always wor­ried that mem­ory was a zero sum game, and that by us­ing a mem­o­riza­tion tool like this I am ar­tifi­cially push­ing out in­for­ma­tion my brain was sav­ing based on its util­ity. I have the same worry about read­ing C.S. or math books that I of­ten don’t end up us­ing im­me­di­ately in my work. Is there any ev­i­dence you’ve seen that could ad­dress this?

Is this analo­gous to the stance-de­pen­dency of agents and in­tel­li­gence?

Fair. (Apart from the bit about hav­ing them si­mul­ta­neously.) I didn’t think of that be­cause I wouldn’t gen­er­ally eat toast with noth­ing on it but but­ter.

I’m in the UK. Dairy prod­ucts here are com­monly pas­teur­ized, but to me UHT means some­thing much more ex­treme which spoils the flavour and I cer­tainly wouldn’t ex­pect cream to be UHT-ed. Is cream re­ally UHT by de­fault in the US? Ewww.

Seems rea­son­able. It also seems rea­son­able to pre­dict oth­ers’ fu­ture ac­tions based on BOTH some­one’s in­ten­tions and their abil­ity to un­der­stand con­se­quences. You may not be able to sep­a­rate these—af­ter the third time some­one yells “FIRE” and runs away, you don’t re­ally know or care if they’re try­ing to cause trou­ble or if they’re just mis­taken about the re­sults.

To me ter­minol­ogy like “puz­zle” seems to sug­gest it is a search for an an­swer but the pro­cess seems also be char­ac­ter­ised by avoidance of in­for­ma­tion gen­er­a­tion.

You could have a challenge of lift­ing a weigth and one could strug­gle by pul­ling or press­ing hard with their mus­cles. “tin­ker­ing” seems to re­fer to cog­ni­tive adap­ta­tion so weightlift­ing doesn’t fit into the defi­ni­tion. But to me it seems it is more about suc­cess rather than smart­ing up. If one phrases it as “I feel un­com­fortable when X hap­pens, let’s do some­thing differ­ent” and “Now I feel com­fortable” it is a challenge and a strug­gle but not a ques­tion or a puz­zle. If one were to ask “What I could do to make my­self com­fortable?” that could be an­swered with knowl­edge or knowl­edge gen­er­a­tion. But it doesn’t seem clear to me whether the strug­gle ac­tu­ally has ques­tion struc­ture.

At most ex­treme it would not be to­tally crazy to de­scribe a weightlifter as an­swer­ing the ques­tion “How do I lift these weights?” and the an­swer be­ing “give mus­cle mo­tor com­mands in the or­der x, y ,z”. I guess some­body could help with weigth­lift­ing with turn­ing it into a puz­zle “hey I see your tech­nique is wrong. Try lift­ing like this.”. But more usu­ally it is a challenge of both­er­ing the effort and maybe liv­ing throught the un­com­forta­bil­ity of the lift. And while even those could be turned into emo­tional in­tel­li­gence ques­tions (“emo­tional tech­nique”) they are not stan­dardly tack­led as ques­tions.

Some­one that is in­ter­ested in “in­stru­men­tal episte­mol­ogy” should be in­ter­ested in in­stru­men­tal any­thing and suc­ceed­ing at a task of­ten in­volves suc­ced­ing in di­men­sions other than episte­mol­ogy too. All mod­els are wrong but some are use­ful so in some situ­a­tions it might be easy to find mod­els that are very use­ful but very sim­ple. Like be­ing a re­li­gious zealot might give a lot of con­fi­dence which could be very use­ful so a con­se­quen­tial­ist mind might recog­nise the suc­cess and lean into that di­rec­tion. Is such an in­duc­tive in­fer­ence rea­son­able? Maybe do­ing quan­tum me­chan­ics as a bind fate black box leads to “shut up and calcu­late” be a more suc­ces­full strat­egy than try­ing to form a bro­ken un­der­stad­ing/​in­tu­ition and suffer many mis­takes. Thus com­pe­tence might mean ab­strac­tion su­pres­sion.

Where are you? In the US cream is gen­er­ally UHT pas­teur­ized, but if you’re some­where where that’s not com­mon your cream won’t last as long.

I’m not sure there’s any pur­pose for which choco­late and ba­con are both suit­able re­place­ments for but­ter.)

Both would be tasty on toast. Even si­mul­ta­neously!

Roam might be great for writ­ing pa­pers etc, but is it a long-term solu­tion for note tak­ing? Who owns your data? What hap­pens when the com­pany goes away?

See­ing you write about this prob­lem, in such harsh terms as “formerly-known-as-ra­tio­nal­ity com­mu­nity” and “effects are iffier and get­ting worse”, is sur­pris­ing in a good way.

Maybe talk­ing clearly could help against these effects. The Amer­i­can talk­ing style has been get­ting more oblique lately, and it’s es­pe­cially bad on LW, maybe due to all the mind prac­tices. I feel this, I guess that, I’d like to un­der­stand bet­ter… For con­trast, read DeMille’s in­ter­view af­ter he quit di­a­net­ics. It’s such a re­fresh­ingly di­rect style, like he spent years mired in oblique talk and mind prac­tices then got fed up and flipped to the op­po­site, to­tal clar­ity. I’d love to see more of that here.

My first im­pres­sion was that this is much too ob­vi­ous to be worth talk­ing about, but my sec­ond thought is that I’ve found it very use­ful to have these lan­guage based trig­gers that act as a “sum­mon sapi­ence” spell. Trig­gers that make you stop and think. They don’t push you to make any par­tic­u­lar choice, but just to no­tice that this is a situ­a­tion where there is a choice to be made.

I won­der if it would be worth putting to­gether a list of “words and phrases which, when you hear them, should make you stop and think”. “Why does this keep hap­pen­ing?” be­longs on that list for sure.

I think Anki is great at learn­ing spe­cific facts, even quite com­plex ones—I have used it ex­ten­sively to learn lan­guages—but it doesn’t offer any op­por­tu­ni­ties to link ideas to­gether. It’s ba­si­cally an effi­cient method of tak­ing facts—even com­plex facts like “what does this sen­tence mean?” or “what did peo­ple say in this short video clip?”—and putting them suffi­ciently into long-term mem­ory that you can then use them in the real world. This fi­nal step is cru­cial as it al­lows these Anki facts to come al­ive and be­come much richer as they be­come part of rich se­man­tic web.

Anki offers no pos­si­bil­ity of link­ing up and de­vel­op­ing ideas. It’s ba­si­cally a very effi­cient mem­ory de­vice.

I found it a bit con­fus­ing that you first re­ffered to se­lec­tion and con­trol as types of op­ti­miz­ers and then (seem­ingly?) re­placed se­lec­tion by op­ti­miza­tion in the rest of the text.

Con­sider this function

$\text{def foo():}\text{a=0.0}\text{for i in range(10**100)}\text{a=a+1/i}\text{return a>2}$

This is valid code that re­turns True. Shouldn’t this also be valid code?

$\text{def foo():}\text{a=0.0}\text{for i in range(}\infty \text{)}\text{a=a+1/(i**2)}\text{return a>1}$

There are a whole space of “pro­grams” that cant be com­puted di­rectly, but can still be rea­soned about.

What you wrote is good, and not worth chang­ing. But I wanted to men­tion that CDT is even more bonkers than that: the pre­dic­tion can be made in the fu­ture, just as long as there is no causal path to how the pre­dic­tor is pre­dict­ing. In some cases, the pre­dic­tor can even know the ac­tion taken, and still pre­dict in a way that CDT thinks is causally dis­con­nected.

Re­lated, there seems to be a de­cent deal of aca­demic liter­a­ture on in­ten­tion vs. in­ter­pre­ta­tion in Art, though maybe less in news and me­dia.

https://​​www.iep.utm.edu/​​art­in­ter/​​#H1

https://​​en.wikipe­dia.org/​​wiki/​​Autho­rial_intent

Some other semi-re­lated links:

https://​​foun­da­tional-re­search.org/​​should-we-base-moral-judg­ments-on-in­ten­tions-or-out­comes/​​

https://​​en.wikipe­dia.org/​​wiki/​​In­ten­tion_(crim­i­nal_law)

https://​​en.wikipe­dia.org/​​wiki/​​Negligence

https://​​en.wikipe­dia.org/​​wiki/​​Reck­less­ness_(law)

Com­mu­ni­ca­tion should be judged for ex­pected value, not in­ten­tion (by con­se­quen­tial­ists)

If I were to scream “FIRE!” in a crowded the­ater, it could cause a lot of dam­age, even if my in­ten­tion were com­pletely un­re­lated. Per­haps I was re­spond­ing to a de­vi­ous friend who asked, “Would you like more pop­corn? If yes, should ‘FIRE!’”.

Not all speech is pro­tected by the First Amend­ment, in part be­cause speech can be used for ex­pected harm.

One com­mon defense of in­cor­rect pre­dic­tions is to claim that their in­ter­pre­ta­tions weren’t their in­ten­tions. “When I said that the US would fall if X were elected, I didn’t mean it would liter­ally end. I meant more that...” Th­ese kinds of state­ments were dis­cussed at length in Ex­pert Poli­ti­cal Judge­ment.

But this defense rests on the idea that com­mu­ni­ca­tors should be judged on in­ten­tion, rather than ex­pected out­comes. In those cases, it was of­ten clear that many peo­ple in­ter­preted these “ex­perts” as mak­ing fairly spe­cific claims that were later re­jected by their au­thors. I’m sure that much of this could have been pre­dicted. The “ex­perts” of­ten definitely didn’t seem to be go­ing out of their way to be mak­ing their af­ter-the-out­come in­ter­pre­ta­tions clear be­fore-the-out­come.

I think that it’s clear that the in­ten­tion-in­ter­pre­ta­tion dis­tinc­tion is con­sid­ered highly im­por­tant by a lot of peo­ple, so much so as to ar­gue that in­ter­pre­ta­tions, even pre­dictable ones, are less sig­nifi­cant in de­ci­sion mak­ing around speech acts than in­ten­tions. I.E. “The im­por­tant thing is to say what you truly feel, don’t worry about how it will be un­der­stood.”

But for a con­se­quen­tial­ist, this dis­tinc­tion isn’t par­tic­u­larly rele­vant. Speech acts are judged on ex­pected value (and thus ex­pected in­ter­pre­ta­tions), be­cause all acts are judged on ex­pected value. Similarly, I think many con­se­quen­tial­ists would claim that here’s noth­ing meta­phys­i­cally unique about com­mu­ni­ca­tion as op­posed to other ac­tions one could take in the world.

Some po­ten­tial im­pli­ca­tions:

1. Much of com­mu­ni­cat­ing on­line should prob­a­bly be about de­vel­op­ing em­pa­thy for the reader base, and a sense for what read­ers will mis­in­ter­pret, es­pe­cially if such mis­in­ter­pre­ta­tion is com­mon (which it seems to be).

2. Analy­ses of the in­ter­pre­ta­tions of com­mu­ni­ca­tion could be more im­por­tant than anal­y­sis of the in­ten­tions of com­mu­ni­ca­tion. I.E. un­der­stand­ing au­thors and artis­tic works in large part by un­der­stand­ing their effects on their view­ers.

3. It could be very rea­son­able to at­tempt to map non prob­a­bil­is­tic fore­casts into prob­a­bil­is­tic state­ments based on what read­ers would in­ter­pret. Then these fore­casts can be scored us­ing scor­ing rules just like those as reg­u­lar prob­a­bil­is­tic state­ments. This would go some­thing like, “I’m sure that Bernie San­ders will be elected” → “The read­ers of that state­ment seem to think the au­thor ap­ply­ing prob­a­bil­ity 90-95% to the state­ment ‘Bernie San­ders will win’” → a brier/​log score.

Note: Please do not in­ter­pret this state­ment as at­tempt­ing to say any­thing about cen­sor­ship. Cen­sor­ship is a whole differ­ent topic with dis­tinct costs and benefits.

Hmm, in­ter­est­ing. When I buy cream (from a su­per­mar­ket; I guess they are very cau­tious) the date they put on it is gen­er­ally about one week in the fu­ture. I’ve taken their word for it and bought it not too long be­fore I need to use it. I should do some ex­per­i­ments...

Tasti­ness (for me) isn’t a scalar thing. You want differ­ent tastes in differ­ent con­texts. (In some sense choco­late is far tastier than but­ter, but there are many pur­poses for which I would use but­ter and would not con­sider us­ing choco­late. The same is true of ba­con. I’m not sure there’s any pur­pose for which choco­late and ba­con are both suit­able re­place­ments for but­ter.)

En­courage­ment to write the top level post, with offer of at least some help al­though pre­sum­ably peo­ple who are there in Berkeley to see it would be more helpful in many ways. This matches my model of what is hap­pen­ing.