Vladimir_Nesov

• Vladimir_Nesov 13 Aug 2019 16:15 UTC

  4 points

  in reply to: Kaj_Sotala's comment on: Ma­chine Learn­ing Anal­ogy for Meditation

  thoughts be­ing the cause of ac­tions is re­lated to a cen­tral strat­egy of many peo­ple around here

  (It’s a good rea­son to at least wel­come ar­gu­ments against this be­ing the case. If your cen­tral strat­egy is built on a false premise, you should want to know. It might be pointless to ex­pect any use­ful info in this di­rec­tion, but I think it’s healthier to still want to see it emo­tion­ally even when you de­cide that it’s not worth your time to seek it out.)

• Vladimir_Nesov 13 Aug 2019 11:01 UTC

  2 points

  in reply to: John_Maxwell_IV's comment on: Open & Wel­come Thread Au­gust 2019

  Another im­por­tant take­away from this ob­ser­va­tion is that there is no point in re­bal­anc­ing a port­fo­lio of stocks with any reg­u­lar­ity, which makes hand-made stock port­fo­lios al­most as effi­cient (in has­sle and ex­penses) as in­dex ETFs. Re­bal­anc­ing is only use­ful to keep it rea­son­ably di­ver­sified and to get rid of stocks that risk re­duced liquidity. This is how in­dex ETFs fall short of the mark of what makes them a good idea: a bet­ter ETF should stop fol­low­ing a dis­tri­bu­tion of an in­dex and only use an in­dex as a cat­a­logue of liquid stocks. Given how low TERs get in large funds, this doesn’t re­ally mat­ter, and ac­countabil­ity/​reg­u­la­tion is eas­ier with keep­ing to the dis­tri­bu­tion from an in­dex, but smaller funds could get lower TERs by fol­low­ing this strat­egy while re­tain­ing all benefits (ex­cept the cru­cial mar­ket­ing benefit of be­ing able to demon­strate how its perfor­mance keeps up with an in­dex). For the same rea­son, cap weighted in­dex ETFs are worse by be­ing less di­ver­sified (which ac­tu­ally makes some in­dex ETFs that hold a lot of stocks a bad choice), while equal-weight ETFs are worse by re­bal­anc­ing all the time (to the point where they can’t get a low TER at all).

  Aside from that, a very low TER in­dex (that’s not too un­bal­anced due to cap-weight­ing) is more di­ver­sified than a hand-made port­fo­lio with 30 stocks, with­out los­ing ex­pected money, so one can use a bit more lev­er­age with it to get a similar risk pro­file with a bit more ex­pected money (lev­er­aged ETFs are hard to judge, but one could make a port­fo­lio that holds some non-lev­er­aged in­dex ETFs in a role similar to bonds in a con­ser­va­tive al­lo­ca­tion, i.e. as lower-risk part, and some self-con­tained lev­er­aged things in the rest of it).

  There might also be tax benefits to how an in­dex han­dles div­i­dends, get­ting more ex­pected money than the same stocks (not sure if this hap­pens in the US, or how it de­pends on tax brack­ets). Similarly, stocks that pay no div­i­dends might be bet­ter for a hand-made port­fo­lio (and there is less has­sle with re­ceiv­ing/​rein­vest­ing div­i­dends or hav­ing to per­son­ally de­clare taxes for them if they are not au­to­mat­i­cally with­held higher in the chain in your ju­ris­dic­tion).

  you could beat an in­dex fund

  (Re­ply­ing to the phrase, not its ap­par­ent mean­ing in con­text.) All liquid stocks give the same ex­pected money as each other, and the same as all in­dices com­posed of them. Differ­ent dis­tri­bu­tions of stocks will have differ­ent ac­tual out­comes, some greater than oth­ers. So of course one can beat an in­dex in an ac­tual out­come (this will hap­pen ex­actly half the time). A sin­gle lev­er­aged stock gives more ex­pected money than any non-lev­er­aged in­dex fund (or any non-lev­er­aged stock), yet makes a very poor in­vest­ment, which illus­trates that beat­ing an in­dex in ex­pec­ta­tion is also not what any­one’s af­ter.

• Vladimir_Nesov 12 Aug 2019 10:40 UTC

  2 points

  in reply to: mr-hire's comment on: Power Buys You Dis­tance From The Crime

  That’s rele­vant to the ex­am­ple, but not to the ar­gu­ment. Con­sider a hy­po­thet­i­cal Jes­sica less in­ter­ested in con­flict the­ory or a topic other than con­flict the­ory. Also, com­mon knowl­edge doesn’t seem to play a role here, and “doesn’t know about” is a level of taboo that con­tra­dicts the as­sump­tion I posited about the ar­gu­ment from se­lec­tion effect be­ing “well-known”.

• Vladimir_Nesov 12 Aug 2019 9:07 UTC

  2 points

  in reply to: jessicata's comment on: Power Buys You Dis­tance From The Crime

  Would you cor­rect your re­sponse so? (Should you?) If the tar­get au­di­ence tends to act similarly, so would they.

  Aside from that, “How do you ex­plain X?” is re­ally am­bigu­ous and an­chors on well-un­der­stood rather than apt fram­ing. “Does mis­take the­ory ex­plain this case well?” is bet­ter, be­cause you may well use a bad the­ory to think about some­thing while know­ing it’s a bad the­ory for ex­plain­ing it. If it’s the best you can do, at least this way you have gears to work with. Not hav­ing a coun­ter­fac­tu­ally read­ily available good the­ory be­cause it’s taboo and wasn’t de­vel­oped is of course ter­rible, but it’s not a rea­son to em­brace the bad the­ory as cor­rect.

• Vladimir_Nesov 12 Aug 2019 0:27 UTC

  2 points

  in reply to: Zack_M_Davis's comment on: Power Buys You Dis­tance From The Crime

  Is “well-known” good enough here, or do you ac­tu­ally need com­mon knowl­edge?

  There is no need for co­or­di­na­tion or de­pen­dence on what oth­ers think. If you ex­pect your­self to be mis­cal­ibrated, you just fix that. If most peo­ple act this way and ac­cept the ar­gu­ment that con­vinced you, then you ex­pect them to have done the same.

• Vladimir_Nesov 11 Aug 2019 23:15 UTC

  11 points

  in reply to: Zack_M_Davis's comment on: Power Buys You Dis­tance From The Crime

  But are the­o­ries that tend to ex­plode and eat up com­mu­nal re­sources there­fore less likely to be true? If not, then avoid­ing them for the sake of pre­serv­ing com­mu­nal re­sources is a sys­tem­atic dis­tor­tion on the com­mu­nity’s be­liefs.

  Ex­pected in­fre­quent dis­cus­sion of a the­ory shouldn’t lower es­ti­mates of its prob­a­bil­ity. (Does the in­tu­ition that such the­o­ries should be seen as less likely fol­low from most nat­u­ral the­o­ries pre­dict­ing dis­cus­sion of them­selves? Er­ro­neous the­o­riz­ing also pre­dicts that, for ex­am­ple “If this state­ment is cor­rect, it will be the only topic of all fu­ture dis­cus­sions.”)

  In gen­eral, it shouldn’t be pos­si­ble to ex­pect well-known sys­tem­atic dis­tor­tions for any rea­son, be­cause they should’ve been re­cal­ibrated away im­me­di­ately. What not dis­cussing a the­ory should cause is lack of pre­ci­sion (or progress), not sys­tem­atic dis­tor­tion.

• Vladimir_Nesov 9 Aug 2019 9:26 UTC

  2 points

  in reply to: Raemon's comment on: Karma-Change Notifications

  I think up­dates were less fre­quent re­cently (e.g. zero up­dates from last week). This should still hap­pen for some peo­ple, though there is maybe only about a hun­dred users with similar num­ber of an­cient com­ments.

• Vladimir_Nesov 6 Aug 2019 4:01 UTC

  4 points

  on: Com­pilers/​PLs book recom­men­da­tion?

  A good se­lec­tion of top­ics on static anal­y­sis is in

  • F Niel­son, HR Niel­son, C Hankin. Prin­ci­ples of Pro­gram Analysis

  Some pre­req­ui­sites for it and other rele­vant things can be picked up from

  • K Cooper, L Tor­c­zon. Eng­ineer­ing a Compiler

  • HR Niel­son, F Niel­son. Se­man­tics with Ap­pli­ca­tions: An Appetizer

• Vladimir_Nesov 10 Jul 2019 6:26 UTC

  4 points

  in reply to: romeostevensit's comment on: Do bond yield curve in­ver­sions re­ally in­di­cate there is likely to be a re­ces­sion?

  More gen­er­ally, mar­ket timers lose hard.

  Do you mean they make a port­fo­lio that’s too con­ser­va­tive, so that lost money be­comes lost util­ity? Or do they lose in some other way? (This sounds su­perfi­cially similar to claims that there are stock/​fu­tures trad­ing strate­gies that sys­tem­at­i­cally lose money other than on fees or spread, which I think can’t hap­pen be­cause the op­po­site strate­gies would then sys­tem­at­i­cally make money.)

• Vladimir_Nesov 9 Jul 2019 4:02 UTC

  2 points

  in reply to: Dagon's comment on: Black hole narratives

  From the other side, agree­ment is of­ten not real. Peo­ple agree out of po­lite­ness, or agree with a dis­torted (per­haps more salient but less rele­vant to the dis­cus­sion) ver­sion of a claim, with­out en­sur­ing it’s not a bucket er­ror. There’s some use in keep­ing be­liefs un­changed, but not in failing to un­der­stand the mean­ing of the claim un­der dis­cus­sion (when it has one). So agree­ment (es­pe­cially your own, as it’s eas­ier to fix) should be treated with scep­ti­cism.

• Vladimir_Nesov 5 Jul 2019 5:55 UTC

  3 points

  on: What are good re­sources for learn­ing func­tional pro­gram­ming?

  My im­pres­sion is that there are no books that even jointly give a rea­son­able se­lec­tion of what you’d need to mas­ter the main ideas in (typed) func­tional pro­gram­ming. Most of this stuff is in pa­pers.

• Vladimir_Nesov 5 Jul 2019 0:53 UTC

  6 points

  in reply to: Ruby's comment on: Causal Real­ity vs So­cial Reality

  Upon re­flec­tion, my an­swer is that I do en­dorse it on many occasions

  The salient ques­tion is whether it’s a good idea to re­spond to pos­si­ble at­tacks in a di­rect fash­ion. Si­tu­a­tions that can be clas­sified as at­tacks (es­pe­cially in a sense that al­lows the at­tacker to re­main un­aware of this fact) are much more com­mon.

• Vladimir_Nesov 4 Jul 2019 22:06 UTC

  35 points

  in reply to: Said Achmiz's comment on: Causal Real­ity vs So­cial Reality

  But if you don’t en­dorse this re­ac­tion—then deal with it your­self.

  I agree with the above two com­ments (Vaniver’s and yours) ex­cept for a cer­tain con­no­ta­tion of this point. Re­jec­tion of own defen­sive­ness does not im­ply en­dorse­ment of in­sen­si­tivity to tone. I’ve been mak­ing this er­ror in mod­el­ing oth­ers un­til re­cently, and I cur­rently cringe at many of my “com­bat­ive” com­ments and fo­rum policy sug­ges­tions from be­fore 2014 or so. In most cases defen­sive­ness is flat wrong, but so is not op­ti­miz­ing to­wards keep­ing the con­ver­sa­tion com­fortable. It’s tempt­ing to shirk that re­spon­si­bil­ity in the name of avoid­ing the dan­ger of com­pro­mis­ing the sig­nal with po­lite dis­tor­tions. But there is a lot of room for safe op­ti­miza­tion in that di­rec­tion, and mak­ing sure peo­ple are aware of this is im­por­tant. “Deal with it your­self” sug­gests ex­clud­ing this pres­sure. Ten years ago, I would have benefit­ted from it.

• Vladimir_Nesov 26 Jun 2019 15:26 UTC

  9 points

  in reply to: Zack_M_Davis's comment on: What does the word “col­lab­o­ra­tive” mean in the phrase “col­lab­o­ra­tive truth­seek­ing”?

  Here’s the link, found from my ac­count by search­ing for “col­lab­o­ra­tive truth­seek­ing”. There is a “Posts from Any­one/​You/​Your Friends” ra­dio con­trol on the left of the search page, so should prob­a­bly work on your own posts as well.

• Vladimir_Nesov 22 Jun 2019 21:21 UTC

  6 points

  in reply to: avturchin's comment on: Up­com­ing sta­bil­ity of values

  Th­ese “meta-val­ues” you men­tion are just val­ues ap­plied to ap­praisal of val­ues. So in these terms it’s pos­si­ble to have val­ues about meta-val­ues and to value change in meta-val­ues. Value drift be­comes in­stru­men­tally un­de­sir­able with greater power over the things you value, and this ar­gu­ment against value drift is not par­tic­u­larly sen­si­tive to your val­ues (or “meta-val­ues”). Even if you pre­fer for your val­ues to change ac­cord­ing to their “nat­u­ral evolu­tion”, it’s still use­ful for them not to change. For change in val­ues to be a good de­ci­sion, you need to see value drift as more ter­mi­nally valuable than its op­por­tu­nity cost (de­crease in value of the fu­ture ac­cord­ing to your pre­sent val­ues, in the event that your pre­sent val­ues un­dergo value drift).

• Vladimir_Nesov 21 Jun 2019 0:42 UTC

  5 points

  on: Let Values Drift

  The strongest ar­gu­ment against value drift (mean­ing the kind of change in cur­rent val­ues that in­volves change in ideal­ized val­ues) is in­stru­men­tal use­ful­ness of fu­ture val­ues that pur­sue ideal­ized pre­sent val­ues. This says noth­ing about ter­mi­nal value of value drift, and a pri­ori we should ex­pect that peo­ple hold pres­ence of value drift as a ter­mi­nal value, be­cause there is no rea­son for the hap­haz­ard hu­man val­ues to sin­gle out the pos­si­bil­ity of zero value drift as most valuable. Value drift is just an­other thing that hap­pens in the world, like kit­tens. Of course valuable value drift must ob­serve proper form even as it breaks ideal­ized val­ues, since most changes are not im­prove­ments.

  The in­stru­men­tal ar­gu­ment is not that strong when your own per­sonal fu­ture val­ues don’t hap­pen to con­trol the world. So the ar­gu­ment still ap­plies to AIs that have sig­nifi­cant in­fluence over what hap­pens in the fu­ture, but not to or­di­nary peo­ple, es­pe­cially not to peo­ple whose val­ues are not par­tic­u­larly un­usual.

• Vladimir_Nesov 11 Jun 2019 23:21 UTC

  2 points

  in reply to: Raemon's comment on: The Schel­ling Choice is “Rab­bit”, not “Stag”

  Thanks!

• Vladimir_Nesov 11 Jun 2019 23:12 UTC

  6 points

  in reply to: Raemon's comment on: The Schel­ling Choice is “Rab­bit”, not “Stag”

  Sure, any re­source would do (like cop­per coins needed to pay for main­te­nance of equip­ment), or just generic “units of re­sources”. My con­cern is that the term “util­ity” is used in­cor­rectly in an oth­er­wise ex­cel­lent post that is tan­gen­tially re­lated to the topic where its tech­ni­cal mean­ing mat­ters, po­ten­tially prop­a­gat­ing this pop­u­lar mis­read­ing.

• Vladimir_Nesov 11 Jun 2019 21:33 UTC

  6 points

  on: The Schel­ling Choice is “Rab­bit”, not “Stag”

  Utility is not a re­source. In the usual ex­pected util­ity set­ting, util­ity func­tions don’t care about af­fine trans­for­ma­tions (all de­ci­sions re­main the same), for ex­am­ple de­creas­ing all util­ities by 1000 doesn’t change any­thing, and there is no sig­nifi­cance to util­ity be­ing pos­i­tive vs. nega­tive. So re­quire­ments to have at least five units of util­ity in or­der to play a round of a game shouldn’t make sense.

  In this post, there is a re­source whose util­ity could maybe pos­si­bly be given by the iden­tity func­tion. But in that case it could also be given by the “times two minus 87” util­ity func­tion and lead to the same de­ci­sions. It’s not re­ally clear that util­ity is iden­tity, since a ⁵⁰⁄₅₀ lot­tery be­tween end­ing up with zero or ten units of the re­source seems much worse than the cer­tainty of ob­tain­ing five units. (“If you’re Elliott, [zero] is a su­per scary re­sult to imag­ine.”)

• Vladimir_Nesov 9 Jun 2019 14:49 UTC

  13 points

  on: Mis­takes with Con­ser­va­tion of Ex­pected Evidence

  To see why it is wrong in gen­eral, con­sider an ex­treme case: a uni­ver­sal law, which you mostly already be­lieve to be true.

  I feel that this ex­am­ple might cre­ate an­other mis­con­cep­tion, that cer­tainty usu­ally begets greater cer­tainty. So here’s an op­pos­ing ex­am­ple, where high con­fi­dence in a be­lief co­ex­ists with high prob­a­bil­ity that it’s go­ing to get less and less cer­tain. You are at a bus stop, and you be­lieve that you’ll get to your des­ti­na­tion on time, as busses here are usu­ally re­li­able, though they only ar­rive once an hour and can de­vi­ate from sched­ule by sev­eral min­utes. If you see a bus, you’ll prob­a­bly ar­rive in time, but ev­ery minute that you see no bus, it’s ev­i­dence of some­thing hav­ing gone wrong, so that the bus won’t ar­rive at all in the next hour. You ex­pect a highly cer­tain be­lief (that you’ll ar­rive on time) to de­crease (a lit­tle bit), which is bal­anced by an un­likely al­ter­na­tive (for each given minute of wait­ing) of it go­ing in the di­rec­tion of greater cer­tainty (if the bus does ar­rive within that minute).