habryka(Oliver Habryka)

• habryka 4 Aug 2020 22:23 UTC

  4 points

  on: My pa­per was sig­nal­ling the whole time—Robin Han­son wins again

  Mod note: Re­placed the link to the image with the ac­tual image, since I as­signed sub­stan­tial prob­a­bil­ity to that be­ing your in­ten­tion, but our ed­i­tor image han­dling be­ing too con­fus­ing and you giv­ing up. But feel free to re­vert it if it was in­ten­tional.

• habryka 4 Aug 2020 20:14 UTC

  2 points

  in reply to: Dagon’s comment on: Why isn’t there a LessWrong App? Is a “blog-app” sus­tain­able/​use­ful for a com­mu­nity?

  Ah, yeah. I agree that the popovers could use a lot of work on mo­bile. I was think­ing you were refer­ring to some­thing else. In this case I agree with you and am rea­son­ably an­noyed at how we han­dle this my­self.

• habryka 4 Aug 2020 19:07 UTC

  4 points

  in reply to: korin43’s comment on: Why isn’t there a LessWrong App? Is a “blog-app” sus­tain­able/​use­ful for a com­mu­nity?

  I agree with this hy­po­thet­i­cally, but I ac­tu­ally think this mat­ters a lot in prac­tice. Browsers are ac­tu­ally just quite slow at do­ing a lot of things. As a con­crete ex­am­ple, the Face­book web-app (which I use since I don’t want Face­book in­stalled on my phone) is many times slower than their na­tive app, with scrol­ling of­ten be­ing visi­bly slug­gish, and click­ing on but­tons hav­ing a visi­ble de­lay.

  Of course it’s pos­si­ble that the app is just writ­ten less well, but given that the same is true for the Twit­ter web-app, and the apps of ba­si­cally any ma­jor plat­form that I have used both as a na­tive app and on the web (Pin­ter­est, Mes­sen­ger, Google Plus in the old days, Adobe Color, Google Docs), I think it’s rea­son­able to con­clude that na­tive apps feel a lot snap­pier and are gen­er­ally faster across the board, at least on my rel­a­tively weak phone CPU (My phone is about ~5 years old now, so it does tend to strug­gle with a rea­son­able num­ber of web­sites). I think a sub­stan­tial chunk of this is ex­plained by the differ­ent lan­guages, and an­other sub­stan­tial chunk is prob­a­bly the re­sult of the differ­ent ecosys­tems, with the javascript ecosys­tem be­ing gen­er­ally not very perfor­mance friendly.

  Again, I am not say­ing at all that it’s im­pos­si­ble, or even that hard, to make a perfor­mant web-app for some­thing in the refer­ence class of LessWrong. I am just say­ing that if you build an app and don’t make perfor­mance an ex­plicit goal of your we­bapp, and don’t put sub­stan­tial effort and de­sign at­ten­tion into it, then the differ­ence be­tween a na­tive app and a we­bapp will be quite sub­stan­tial. Of course if you do put in that at­ten­tion and thought you can make it snappy in javascript. As you say, fun­da­men­tally you are just putting text in boxes and the the­o­ret­i­cal limit of perfor­mance is far far above what you need for an app like this.

• habryka 4 Aug 2020 18:59 UTC

  2 points

  in reply to: Dagon’s comment on: Why isn’t there a LessWrong App? Is a “blog-app” sus­tain­able/​use­ful for a com­mu­nity?

  And over­ri­den click behaviors

  Huh, do you have any con­crete ex­am­ples of this? I think al­most all of our but­tons that could be links, are in­deed links, and if you no­tice any­thing that is a weird javascript link that takes you to a differ­ent page but you can’t open in a new tab, please let us know. I would con­sider that a straight­for­ward bug, since I also hate links that I can’t open in new tabs.

• habryka 3 Aug 2020 8:06 UTC

  3 points

  in reply to: Pongo’s comment on: Why isn’t there a LessWrong App? Is a “blog-app” sus­tain­able/​use­ful for a com­mu­nity?

  Given that na­tive mo­bile apps are of­ten writ­ten in C, ob­jec­tive-C or C++, just on a sim­ple lan­guage level you should ex­pect some­thing like 3-10x speedup from the switch from an in­ter­preted lan­guage to a com­piled lan­guage (some­times more, some­times less, de­pend­ing on the ap­pli­ca­tion). In gen­eral in­ter­preted lan­guages con­tinue to be sub­stan­tially slower (and though in­ter­preted lan­guages have been get­ting faster due to smarter ex­e­cu­tion en­vi­ron­ments and lots of fancy tricks, so have com­piled lan­guages, due to im­prove­ments in com­pilers).

  The com­par­i­son is pretty similar to the python vs. C com­par­i­son, which I have a good amount of con­crete ex­pe­rience with, since it’s su­per com­mon that if you write ML code, and want to op­ti­mize any spe­cific part of it, you rewrite it in C or C++ and then call that C code from Python. Do­ing so usu­ally gives you some­thing like a 10x perfor­mance im­prove­ment, though it de­pends a bit on ex­actly what you are do­ing. I ex­pect the differ­ence be­tween C++ and javascript to be similar.

  So yes, I do think that pro­gres­sive web-apps run­ning on javascript does mean they are fun­da­men­tally slower than na­tive apps writ­ten in C, C++, or even Swift or Java. This doesn’t mean it’s im­pos­si­ble to write a perfor­mant pro­gres­sive web-app, but it does likely mean that you have to put a lot more thought into op­ti­miza­tion than you would if you were to write na­tive code (also be­cause the javascript ecosys­tem re­ally isn’t well-op­ti­mized for perfor­mance, and so it’s re­ally easy to build a non-perfor­mant app).

• habryka 3 Aug 2020 0:29 UTC

  12 points

  in reply to: gbear605’s comment on: Why isn’t there a LessWrong App? Is a “blog-app” sus­tain­able/​use­ful for a com­mu­nity?

  This seems roughly right to me. We’ve tried rea­son­ably hard (though we could definitely do bet­ter) to make the mo­bile ver­sion of the site work well enough to use it on phones, which I do think is im­por­tant, but I don’t feel like on the mar­gin mak­ing a na­tive app ver­sion is that valuable. In the long run I do hope that we can make LessWrong into a proper pro­gres­sive web-app, which would al­low users to read ar­ti­cles offline, ac­cess the site through their home-screen and gen­er­ally treat it like an app (though perfor­mance wise it would be worse, since pro­gres­sive web-apps run on Javas­ript and na­tive apps get to be writ­ten in much faster lan­guages).

  There is also one ad­di­tional con­sid­er­a­tion, which is that I am wor­ried about mak­ing LessWrong the kind of thing that is pri­mar­ily a phone app. I think by their na­ture phones are much worse for longform con­tent, both read­ing and cre­at­ing it, and I think a LessWrong that was pre­dom­i­nantly used by peo­ple on their phones would be forced to have much shorter con­tent, and cor­re­spond­ingly be a lot more like the rest of the in­ter­net in it’s pres­sure to be short and snappy and as a re­sult of that fail to be able to grap­ple with prob­lems in any real depth. I would never write a full LessWrong post on my phone, and even no­tice that the com­ments that I write on my phone or Ipad tend to be lower qual­ity, with fewer links to ex­ter­nal re­sources, and less thought put into the for­mat­ting or con­tent. It’s also much harder to pre­sent a lot of in­for­ma­tion on a small phone screen, mak­ing cer­tain types of in­tel­lec­tual la­bor quite difficult (as an ex­am­ple, if we were de­vel­op­ing our tag­ging sys­tem pri­mar­ily for a phone-driven au­di­ence, I think it would have to be quite a lot sim­ple and lose a lot of func­tion­al­ity, or re­quire a lot more de­sign effort put into it, which then cor­re­sponds to us build­ing fewer other things).

  I do think it’s likely that if things go well, we will even­tu­ally have an app, but these trade­offs are why it hasn’t been a pri­or­ity till now, and prob­a­bly won’t be for a while longer.

• habryka 2 Aug 2020 4:42 UTC

  4 points

  in reply to: Pattern’s comment on: Power as Easily Ex­ploitable Opportunities

  Seems plau­si­ble to me, though I am not sure whether I would say prob­a­ble. If you could do some­thing like back­prop through the hu­man mind, then I think I can imag­ine out­comes similar to this:

  https://​​youtu.be/​​u5wtoH0_KuA?t=227

• habryka 2 Aug 2020 0:21 UTC

  2 points

  in reply to: Ruby’s comment on: Open Call for Taggers

  Yeah, ok. I do think per­son­al­iza­tion is blocked on Al­go­lia, and I didn’t re­ally think about this as a po­ten­tial solu­tion to this (but it to­tally is). So yeah, maybe slight­ing Al­go­lia was the right call.

• habryka 1 Aug 2020 21:25 UTC

  3 points

  in reply to: Ruby’s comment on: Open Call for Taggers

  I am pretty con­fused what any of this has to do with Al­go­lia. The pri­mary prob­lem to me ap­pears to be that we don’t ac­tu­ally have a large frac­tion of the tags cat­e­go­rized in the tag hi­er­ar­chy dis­played on the All Tags page. We could show you a copy of the tag page table, but that would omit a lot of new tags, and also prob­a­bly not be dense enough. We could de­velop some cus­tom UI for that menu to group them, but that’s mostly a bunch of work (and doesn’t have su­per much to do with Al­go­lia).

  The site search will prob­a­bly always have some­what differ­ent con­straints than nor­mal database op­er­a­tions (in par­tic­u­lar if we want to stay within the au­to­com­plete paradigm), so I don’t think any­thing about this would get eas­ier if we switch away from Al­go­lia (things like this are ac­tu­ally a do­main where Al­go­lia is pretty great).

• habryka 1 Aug 2020 19:15 UTC

  5 points

  in reply to: Andy Jones’s comment on: Are we in an AI over­hang?

  Out of cu­ri­os­ity, what was it that con­vinced you this isn’t an in­fo­haz­ard-like risk?

  Some mix­ture of:

  • I think it’s pretty valuable to have open con­ver­sa­tion about be­ing in an over­hang, and I think on the mar­gin it will make those wor­lds go bet­ter by im­prov­ing co­or­di­na­tion. My cur­rent sense is that the per­spec­tive pre­sented in this post is rea­son­ably fre­quent among peo­ple in ML, so that marginally re­duc­ing how many peo­ple be­lieve this is not go­ing to do much of a differ­ence, but hav­ing good write­ups that sum­ma­rize the ar­gu­ments seems like it has a bet­ter chance of cre­at­ing some kind of com­mon knowl­edge that al­lows peo­ple to co­or­di­nate bet­ter here.

  • This post more so than other posts in its refer­ence class em­pha­sizes a bunch of the safety con­cerns, whereas I ex­pect the next post to re­place it to not do that very much

  • Cu­ra­tion in par­tic­u­lar mostly sends out the post to more peo­ple who are con­cerned with safety. This post found a lot of trac­tion on HN and other places, so in some sense the cat is out of the bag and if it was harm­ful the cu­ra­tion de­ci­sion won’t change that very much, and it seems like it would un­nec­es­sar­ily hin­der the peo­ple most con­cerned about safety if we don’t cu­rate it (since the con­sid­er­a­tions do also seem quite rele­vant to safety work).

• habryka 1 Aug 2020 19:04 UTC

  2 points

  on: Fore­cast­ing Newslet­ter: July 2020.

  Mod note: Fixed the bro­ken for­mat­ting. Looks like you pasted some mark­down into our WYSIWYG ed­i­tor.

• habryka 1 Aug 2020 6:19 UTC

  LW: 5 AF: 3

  AF

  on: Are we in an AI over­hang?

  Pro­moted to cu­rated: I think the ques­tion of whether we are in an AI over­hang is pretty ob­vi­ously rele­vant to a lot of think­ing about AI Risk, and this post cov­ers the topic quite well. I par­tic­u­larly liked the use of a lot of small fermi es­ti­mate, and how it cov­ered a lot of ground in rel­a­tively lit­tle writ­ing.

  I also re­ally ap­pre­ci­ated the dis­cus­sion in the com­ments, and felt that Gw­ern’s com­ment on AI de­vel­op­ment strate­gies in par­tic­u­lar help me build a much map of the mod­ern ML space (though I wouldn’t want it to be in­ter­preted as a com­plete map of a space, just a kind of foothold that helped me get a bet­ter grasp on think­ing about this).

  Most of my im­me­di­ate cri­tiques are for­mat­ting re­lated. I feel like the listed sec­tion could have used some more clar­ity, maybe by bold­ing the name for each bul­let point con­sid­er­a­tion, but it flowed pretty well as is. I was also a bit con­cerned about there be­ing some in­fo­haz­ard-like risks from pro­mot­ing the idea of be­ing in an AI over­hang too much, but af­ter talk­ing to some more peo­ple about it, and think­ing for a bit about it, de­cided that I don’t think this post adds much ad­di­tional risk (e.g. by en­courag­ing AI com­pa­nies to act on be­ing in an over­hang and try to dras­ti­cally scale up mod­els with­out con­cern for safety).

• habryka 31 Jul 2020 17:56 UTC

  15 points

  in reply to: Yoav Ravid’s comment on: Open Call for Taggers

  Ahh, hmm. That is em­bar­rass­ing. Hmm, I wish I had a bet­ter ex­cuse for this. Hmm…

  I mean, look over there a three-headed mon­key!

  (Will be fixed within the hour)

  Edit: And it’s fixed. Sorry about that!

• habryka 31 Jul 2020 3:31 UTC

  5 points

  in reply to: mr-hire’s comment on: PSA: Tag­ging is Awesome

  And here is me ask­ing it to ex­tend the list of all Ra­tion­al­ity tags:

  -------- My In­put -------
  Alief
  An­ti­ci­pated Ex­pe­riences
  Bayes The­o­rem /​ Bayesi­anism
  Bounded Ra­tion­al­ity
  Cached Thoughts
  Con­ser­va­tion of Ex­pected Ev­i­dence
  De­ci­sion The­ory
  Episte­mol­ogy
  Game The­ory
  Han­so­nian Pre-Ra­tion­al­ity
  Law-Think­ing
  New­comb’s Prob­lem
  Ro­bust Agents
  Solomonoff In­duc­tion
  Utility Func­tions
  Value of In­for­ma­tion
  Bet­ting
  Dark Arts
  Fo­cus­ing
  Fore­cast­ing & Pre­dic­tion
  Goal Fac­tor­ing
  Ham­ming Ques­tions
  Iden­tity
  In­side/​Out­side View
  In­ter­nal Dou­ble Crux
  In­tro­spec­tion
  Notic­ing
  Schol­ar­ship & Learn­ing
  Tech­niques
  Trig­ger-Ac­tion Plan­ning
  Affect Heuris­tic
  Bucket Er­rors
  Com­part­men­tal­iza­tion
  Con­fir­ma­tion Bias
  Fal­la­cies
  Good­hart’s Law
  Group­think
  Heuris­tics and Bi­ases
  Mind Pro­jec­tion Fal­lacy
  Mo­ti­vated Rea­son­ing
  Pica
  Pit­falls of Ra­tion­al­ity
  Ra­tion­al­iza­tion
  Self-De­cep­tion
  Sunk-Cost Fal­lacy
  Com­mon Knowl­edge
  Con­ver­sa­tion
  De­cou­pling vs Con­tex­tu­al­iz­ing
  Disagree­ment
  In­fer­en­tial Dis­tance
  In­for­ma­tion Cas­cades
  Philos­o­phy of Lan­guage
  Steel­man­ning
  Con­scious­ness
  Dual Pro­cess The­ory (Sys­tem 1 & 2
  Gen­eral In­tel­li­gence
  Subagents
  Pre­dic­tive Pro­cess­ing
  Per­cep­tual Con­trol The­ory
  Cen­ter for Ap­plied Ra­tion­al­ity
  Cu­ri­os­ity
  Gears-Level
  Memetic Im­mune Sys­tem
  Prac­tice & Philos­o­phy of Science
  Ra­tion­al­ity Quotes
  
  ------- GPT3′s Re­sponse -------
  
  Men­tal Si­mu­la­tion
  Mind Pro­jec­tion Fal­lacy
  Mo­dal Ra­tion­al­ity
  Nar­ra­tive Causal­ity
  Si­mu­la­tion Heuris­tic
  Sub­junc­tive Prob­a­bil­ity
  Bayes’ The­o­rem
  Blind Spots
  Coun­ter­fac­tu­als
  Deon­tol­ogy
  Deon­tol­ogy Quotes
  Epistemic Re­gret
  Epistemic Utility
  Eter­nal­ism
  Eter­nal­ism Quotes
  Ex­pe­ri­en­tial­ism
  Fic­tion­al­ism
  Fic­tion­al­ism Quotes
  Game The­ory
  Han­so­nian Pre-Ra­tion­al­ity
  Law-Think­ing
  Log­i­cal Fal­la­cies
  Pre­dic­tive Pro­cess­ing
  Pre­dic­tive Pro­cess­ing Quotes
  Psy­chol­ogy of Science
  Ra­tion­al­ity
  Ra­tion­al­ity Quotes
  Science & Ra­tion­al­ity
  Science Quotes
  Scien­tific Progress
  Scien­tific Progress Quotes
  Scien­tific Real­ism
  Scien­tific Real­ism Quotes
  Solomonoff’s The­ory
  Solomonoff’s The­ory Quotes
  Solomonoff’s Prior
  Solomonoff’s Prior Quotes
  Solomonoff’s In­duc­tion
  Solomonoff’s In­duc­tion Quotes
  Solomonoff’s Al­gorithm
  Solomonoff’s Al­gorithm Quotes
  Ra­tion­al­ist Com­mu­nity
  Affec­tive Death Spirals
  An­thropic Bias
  Bayes’ The­o­rem
  Bias Heuris­tic
  Con­fir­ma­tion Bias
  De­duc­tive vs In­duc­tive
  Epistemic Sta­tus
  Epistemic Utility
  Ex­pert Tes­ti­mony
  Ex­trap­o­la­tive Ten­den­cies
  False-Pos­i­tive Paradox

• habryka 31 Jul 2020 3:27 UTC

  4 points

  in reply to: mr-hire’s comment on: PSA: Tag­ging is Awesome

  I just did it again, ask­ing it to ex­tend the list of all the AI tags.

  ------- My In­put -----
  
  AIXI(23)
  Com­plex­ity of Value(8)
  Cor­rigi­bil­ity(12)
  De­ci­sion The­ory(83)
  Embed­ded Agency(13)
  Fixed Point The­o­rems(7)
  Good­hart’s Law(29)
  In­ner Align­ment(17)
  In­stru­men­tal Con­ver­gence(9)
  Log­i­cal In­duc­tion(22)
  Mesa-Op­ti­miza­tion(10)
  My­opia(6)
  New­comb’s Prob­lem(4)
  Op­ti­miza­tion(21)
  Orthog­o­nal­ity Th­e­sis(4)
  Outer Align­ment(7)
  Solomonoff In­duc­tion(9)
  Utility Func­tions(20)
  AI Box­ing (Con­tain­ment)(8)
  De­bate (AI safety tech­nique)(9)
  Fac­tored Cog­ni­tion(8)
  Hu­mans Con­sult­ing HCH(7)
  Im­pact Mea­sures(21)
  In­verse Re­in­force­ment Learn­ing(14)
  Iter­ated Am­plifi­ca­tion (26)
  Trans­parency /​ In­ter­pretabil­ity(9)
  Value Learn­ing(39)
  AI Risk(53)
  AI Ser­vices (CAIS)(8)
  AI Take­off(38)
  AI Timelines(28)
  CHAI (UC Berkeley)(12)
  FHI (Oxford)(13)
  MIRI(72)
  OpenAI(17)
  Ought(7)
  Alpha- (al­gorithm fam­ily)(10)
  GPT(44)
  Re­search Agen­das(27)
  
  -------- GPT3′s Re­sponse ---------
  
  Hap­piness /​ Life Satis­fac­tion(3)
  Ex­is­ten­tial Risks(6)
  Growth Mind­set(6)
  Op­ti­miza­tion(21)
  Me­tacog­ni­tion(9)
  Prob­a­bil­is­tic Think­ing(9)
  Ra­tion­al­ity(33)
  Retro­duc­tion(9)
  Semi-HUMANs(6)
  Su­per­in­tel­li­gence(40)
  Tractabil­ity(18)
  Trust(5)
  VNM(44)
  Books(1)
  Causal Ex­pla­na­tion(5)
  Com­pu­ta­tional Equiv­alence(9)
  Com­pu­ta­tional Models(13)
  Com­pu­ta­tion­al­ism(11)
  Com­pu­ta­tional Psy­choso­ciol­ogy(8)
  Com­pu­ta­tional So­ciol­ogy(8)
  Con­fer­ences(4)
  Con­trol The­ory(6)
  De­liber­a­tion(10)
  Egal­i­tar­i­anism(7)
  Eth­i­cal Align­ment(7)
  Eth­i­cal Uncer­tainty(6)
  Evolu­tion­ary Game The­ory(5)
  Ex­per­i­men­tal Ethics(9)
  Ex­per­i­men­tal Philos­o­phy(9)
  Fair­ness(5)
  Futarchy(8)
  Game The­ory(9)
  Ge­netic Al­gorithms(6)
  Hap­piness /​ Life Satis­fac­tion(3)
  Hu­man Level AI(7)
  Hu­man-level In­tel­li­gence(5)
  Hu­man-Level Sys­tems(5)
  Im­pact Mea­sures(21)
  In­cen­tives(6)

• habryka 29 Jul 2020 22:28 UTC

  6 points

  in reply to: ESRogs’s comment on: Open Call for Taggers

  Now ac­tu­ally fixed (there was a typo in the URL).

• habryka 29 Jul 2020 19:50 UTC

  2 points

  in reply to: DanielFilan’s comment on: Fo­rum par­ti­ci­pa­tion as a re­search strategy

  Yeah, and I ex­per­i­mented a bunch with that (di­rectly turn­ing fo­rum dis­cus­sions into posts) and mostly felt like it didn’t re­ally work that well. I mostly up­dated that there needs to be a larger syn­the­sis step, though I still have some guesses for more di­rect things that could work. Ben spent some hours dis­till­ing the dis­cus­sion and com­ments on a bunch of posts, which we should get around to post­ing (I just re­al­ized we never pub­lished them).

  Re tag­ging: In gen­eral the tag­ging sys­tem that we are build­ing has a lot in com­mon with be­ing a wiki (col­lab­o­ra­tively ed­itable de­scrip­tions, pro­vid­ing canon­i­cal defi­ni­tions and refer­ences, and pro­vid­ing good sum­maries of ex­ist­ing con­tent), and I ex­pect it to grow into be­ing more of a wiki over time (the tag­ging use-case was a spe­cific nar­row use-case that seemed easy to get trac­tion on, but the mid-term goal is to do a lot more wiki-like stuff). And I think from that per­spec­tive it’s more clear how it helps with dis­til­la­tion.

• habryka 29 Jul 2020 19:34 UTC

  7 points

  in reply to: ESRogs’s comment on: Open Call for Taggers

  I think tag­ging is ac­tu­ally pretty hard. Like, by de­fault you get a ton of syn­onyms of the same con­cepts, and there aren’t good redi­rects, and the tags don’t have good de­scrip­tions, and there is lots of am­bi­guity, and when some­one cre­ates a new tag old posts don’t re­li­ably get tagged. Our tag­ging sys­tem is also more similar to be­ing a wiki, and in-gen­eral my re­search into wikis sug­gests that ba­si­cally all func­tional ones are main­tained by a rel­a­tively small group of highly ded­i­cated ed­i­tors, and that it gen­er­ally doesn’t work to just have ev­ery­one ran­domly edit and add things.

• habryka 29 Jul 2020 17:10 UTC

  8 points

  in reply to: Kaj_Sotala’s comment on: Open Call for Taggers

  Note: This is most viewed for the last 30 days I think. That post in par­tic­u­lar tends to show up at the top ev­ery few months when­ever you get closer to U.S. Col­lege ap­pli­ca­tion pe­ri­ods.

  And yeah, some­times the posts that get a ton of views re­ally aren’t very good. For a while one of our most viewed post was one at −4 karma called “The Effects of Reli­gion [Draft]”. It‘s been one of the rea­sons why I’ve been hes­i­tant to in­clude views as an eas­ily ac­cessible met­ric on the site, be­cause I know how fre­quently it di­verges from qual­ity.

• habryka 29 Jul 2020 17:02 UTC

  2 points

  in reply to: DanielFilan’s comment on: Fo­rum par­ti­ci­pa­tion as a re­search strategy

  This is ac­tu­ally a ma­jor mo­ti­va­tion for the wiki/​tag­ging sys­tem we are build­ing. Also, you might have no­ticed all the ed­ited tran­scripts we’ve been pub­lish­ing, and the de­bates we’ve started or­ga­niz­ing, which is also part of this. I‘be ex­per­i­mented a lot over the last year with UI for po­ten­tially di­rectly dis­till­ing com­ment threads, but all of them ended up too clunky and messy to ever make me ex­cited about them, though I still have some things I might want to give a shot, but over­all I am cur­rently think­ing of tack­ling this prob­lem in a slightly more in­di­rect way.