G Gordon Worley III

• G Gordon Worley III 23 Apr 2019 20:40 UTC

  2 points

  on: Thoughts on Hu­man Models

  So a num­ber of the com­ments have pointed in the di­rec­tion of my con­cerns with what I in­ter­pret to be the un­der­ly­ing as­sump­tion of this post, namely that it is at all pos­si­ble to work with some­thing that is not at least touched by hu­mans enough that im­plicit, par­tial mod­el­ing of hu­mans will not hap­pen as a re­sult of try­ing to build AI, gen­eral or nar­row, even if re­stricted to a small do­main. This is not to fail to ac­knowl­edge that much cur­rent AI safety work tends to be ex­tremely hu­man-cen­tric, go­ing so far as to rely on uniquely hu­man ca­pa­bil­ities (at least unique among known things), and that this is in it­self a prob­lem for many of the rea­sons you lay out, but I think it would be a mis­take to think we can some­how get away from hu­mans in build­ing AGI.

  The re­al­ity is that hu­mans are in­volved in the work of build­ing AGI, in­volved in the de­sign and con­struc­tion of the hard­ware they will run on, the data sets they will use, etc., and even if we think we’ve re­moved the la­tent hu­man-shaped pat­terns from our al­gorithms, hard­ware, and data, we should strongly sus­pect we are mis­taken be­cause hu­mans are tremen­dously bad at notic­ing when they are as­sum­ing some­thing true of the world when it is ac­tu­ally true of their un­der­stand­ing, i.e. I would ex­pect it to be more likely that hu­mans would fail to no­tice their la­tent pres­ence in a “hu­man-model-free” AI than for the AI to ac­tu­ally be free of hu­man mod­el­ing.

  Thus to go down the di­rec­tion of work­ing on build­ing AGI with­out hu­man mod­els risks failure be­cause we failed to deal with the AGI pick­ing up on the la­tent pat­terns of hu­man­ity within it. This is not to say that we should stick to a hu­man-cen­tric ap­proach, be­cause it has many prob­lems as you’ve de­scribed, but to try to avoid hu­mans is to ig­nore mak­ing our sys­tems ro­bust to the kinds of in­terfer­ence from hu­mans that can push us away from the goal of safe AI, es­pe­cially un­ex­pected and un­planned for in­terfer­ence due to hid­den hu­man in­fluence. If we in­stead build ex­pect­ing to deal with and be ro­bust to the in­fluence of hu­mans, we stand a much bet­ter chance of pro­duc­ing safe AI than ei­ther be­ing hu­man-cen­tric or overly ig­nor­ing hu­mans.

• G Gordon Worley III 18 Apr 2019 21:15 UTC

  2 points

  in reply to: shminux's comment on: Quan­ti­ta­tive Philos­o­phy: Why Si­mu­late Ideas Numer­i­cally?

  I am not sure I fol­low. The bots in­deed do end up clus­ter­ing into 4 to 5 differ­ent clusters, where each cluster rep­re­sents a cer­tain con­ver­gent view. By “keep­ing the af­finity score”, do you mean they keep track of the past in­ter­ac­tions, not just com­pare cur­rent views at each step? That would be an in­ter­est­ing im­prove­ment, adding mem­ory to the model, but that would be, well, an im­prove­ment, not nec­es­sar­ily some­thing you put into a toy model from the be­gin­ning. Maybe you mean some­thing else? I’m con­fused.

  Oh, this para­graph seems to sug­gest your model has a lot more go­ing on that I got from read­ing this post. Maybe if I fol­lowed you links I would find some more de­tails (sounded like they were just ex­tra de­tails that could be skipped)? I got the im­pres­sion you found a func­tion that has a shape illus­tra­tive of what you want and that was it, but this sounds like there’s a lot more go­ing on not de­scribed in the text of this post!

• G Gordon Worley III 17 Apr 2019 2:17 UTC

  9 points

  on: Quan­ti­ta­tive Philos­o­phy: Why Si­mu­late Ideas Numer­i­cally?

  It’s un­clear to me that the model you con­struct has much re­la­tion­ship to your tested idea, or at least any more than any other words or images you could have pro­vided to illus­trate the idea. I get that you see peaks and valleys in the func­tion that you in­ter­pret as re­lated to at­trac­tion and re­pul­sion of peo­ple to­wards clusters, but again I don’t see any clear ground­ing of this model in the idea you are test­ing, so to me it just feels like you drew a pic­ture to illus­trate the idea in a very con­voluted way. Put an­other way, I don’t feel like your model con­strains your be­lief in the idea at all, be­cause you could have come up with any func­tion to draw any curve you could have wanted to show us.

  Maybe I’ve missed some­thing where you did provide a ground­ing, but I read back through the post a sec­ond time and still didn’t see any­thing defini­tive. If there is some­thing like that, or if you left it out and can ex­plain how the model is causally con­nected to the idea such that some­thing about the model says some­thing about the idea, can you do so now?

  FWIW I went in to this ex­pect­ing a very differ­ent sort of model, one more like a simu­la­tion us­ing sim­ple bots that in­ter­act in sim­plified ways you de­scribe and then we could see how they end up clus­ter­ing, maybe by each bot keep­ing an af­finity score for the oth­ers and find­ing re­sults about the af­finity of the bots form­ing clusters. That feels to me like it does have some more ground­ing in the idea be­ing tested, where each bot is sim­plified stand-in for a per­son and their in­ter­ac­tions and af­finity scores for the other bots stand in for hu­man in­ter­ac­tions and hu­man af­finity for other hu­mans. But again maybe you have some rea­son for think­ing the model you pre­sent has the same kind of causal con­nec­tion by shar­ing some struc­tural similar­ity; I just don’t see it so would ap­pre­ci­ate if you could clar­ify.

• G Gordon Worley III 16 Apr 2019 23:59 UTC

  6 points

  in reply to: Chris_Leong's comment on: Open Prob­lems in Archipelago

  Although I guess there’s also the ques­tion of, why don’t we just cre­ate an archipelago of sub­red­dits on red­dit if that’s the di­rec­tion we want to go? Just prepend the name of each sub­red­dit with “LessWrong” and link them to­gether some­how and be done with it.

  I think we all know the an­swer, though: LW has cer­tain stan­dards and does a bet­ter job of keep­ing out cer­tain kinds of noise than red­dit does, even with ac­tive mod­er­a­tion. LW to­day at­tracts cer­tain folks, de­ters oth­ers, and its bound­aries make it a com­pel­ling gar­den to hang out in, even if not ev­ery­one agrees on, say, whether we should al­low only flow­er­ing plants in our gar­den or if ferns and moss are okay.

  I like the di­rec­tion of hav­ing LW, EA Fo­rum, and Align­ment Fo­rum be­ing semi-con­nected; I would love if EA Fo­rum func­tions more like the Align­ment Fo­rum does in re­la­tion to LW, and I think it would be cool to po­ten­tially see one or two ad­di­tional sites branch off if that made sense, but I also don’t feel like there’s enough vol­ume here that I’d en­joy see­ing us frac­ture too much, be­cause there’s a lot of benefit too in keep­ing things to­gether and ex­pos­ing folks to things they oth­er­wise might not see be­cause it hap­pens to be loosely con­nected enough to the things they do want to see that they end up en­coun­ter­ing it. I en­joy stum­bling on things I had no idea I would learn some­thing from, but oth­ers are less open in this way and have differ­ent prefer­ences.

• G Gordon Worley III 16 Apr 2019 19:10 UTC

  4 points

  on: High­lights from “In­te­gral Spiritu­al­ity”

  The fol­low­ing quote was called out in a deleted com­ment, but I think there is some­thing to dis­cuss here that would be missed if we didn’t come back to it even though that com­ment was ruled off-topic.

  Thus, ul­ti­mate con­cern was dis­placed to sci­ence, a con­cern that its meth­ods were sim­ply not ca­pa­ble of han­dling. And sci­ence it­self was always com­pletely hon­est about its limi­ta­tions: sci­ence can­not say whether God ex­ists or does not ex­ist; whether there is an Ab­solute or not; why we are here, what our ul­ti­mate na­ture is, and so on. Of course sci­ence can find no ev­i­dence for the Ab­solute; nor can it find ev­i­dence dis­prov­ing an Ab­solute. When sci­ence is hon­est, it is thor­oughly ag­nos­tic and thor­oughly quiet on those ul­ti­mate ques­tions.

  The now deleted com­plaint was that this is say­ing some­thing like sci­ence is in a non-over­lap­ping mag­is­terium from the ques­tion of whether or not God ex­ists. I agree try­ing to claim sep­a­rate mag­is­terium is a prob­lem and doesn’t work, so what do I see as the value of in­clud­ing this quote?

  Mainly to high­light a point that I think is of­ten poorly un­der­stood: that sci­ence, for all the good it does, in­ten­tion­ally cuts it­self off from cer­tain kinds of ev­i­dence in or­der to al­low it to func­tion. Maybe we can de­bate what is the “real” sci­ence, but I’m think­ing here of the nor­mal, run-of-the-mill thing you’d call “sci­ence” we find go­ing on in uni­ver­si­ties around the world, and that form of sci­ence speci­fi­cally ig­nores lines of ev­i­dence we might call anec­do­tal or phe­nomenolog­i­cal and, for our pur­poses, ig­nores ques­tions of episte­mol­ogy by set­tling for a kind of episte­molog­i­cal prag­ma­tism that al­lows sci­ence to get on with the busi­ness of sci­ence with­out hav­ing to re­solve philos­o­phy prob­lems ev­ery time you want to pub­lish a pa­per on fruit flies.

  This choice to prag­mat­i­cally ig­nore deep episte­molog­i­cal ques­tions is a good choice for sci­ence, of course, be­cause it lets it get things done, but it also means we can­not take re­sults like “sci­ence finds no ev­i­dence of su­per­nat­u­ral be­ings or some ever-pre­sent unify­ing force we could rea­son­ably la­bel God” as stronger ev­i­dence than it is. Yes, this is pretty strong ev­i­dence that there is no God like the kind you find in a re­li­gious text that in­ter­acts with the world, but it’s also not much ev­i­dence of any­thing about a God that’s more like an in­visi­ble dragon liv­ing in a garage. The thing that lets you ad­dress those sorts of ques­tions is a bit differ­ent from what is typ­i­cally done un­der the ban­ner of sci­ence.

  This quote does go a bit too far when it says sci­ence should be “thor­oughly quiet on those ul­ti­mate ques­tions”, be­cause it does have some­thing to say, but I still thought it worth in­clud­ing be­cause it high­lights the com­mon over­reach of sci­ence into do­mains which it speci­fi­cally rules it­self our from par­ti­ci­pat­ing in by set­ting up its method­olog­i­cal as­sump­tions so that it can func­tion.

  (This last point put an­other way, think of how an­noyed you’d be if ev­ery time you told your friend you felt sad and wanted a hug they said “I don’t know, I can’t re­ally mea­sure your sad­ness very well, and it’s just you re­port­ing this sad­ness any­way, so I can’t tell if it’s worth it to give you the hug”.)

• G Gordon Worley III 16 Apr 2019 18:04 UTC

  9 points

  in reply to: Richard_Kennaway's comment on: Slack Club

  I have two thoughts on this.

  One is that differ­ent spiritual tra­di­tions have their own deep, com­plex sys­tem of jar­gon that some­times stretch back thou­sands of years through mul­ti­ple trans­la­tions, schisms, and acts of syn­cretism. So when you first en­counter it you can feel like it’s a lot and it’s new and why can’t these peo­ple just talk nor­mally.

  Of course, most LW read­ers live in a world full of jar­gon even be­fore you add on the LW jar­gon, much of it from STEM dis­ci­plines. Peo­ple from out­side that cluster feel much the same way about STEM jar­gon as the av­er­age LW reader may feel about spiritual jar­gon. I point this out merely be­cause I re­al­ized, when you brought up the spiritual ex­am­ple, that I wasn’t given a full ac­count of what’s differ­ent about ra­tio­nal­ists, maybe, in that there’s a ten­dency to make new jar­gon even when a liter­a­ture search would re­veal ex­ist­ing jar­gon ex­ists.

  Which is rele­vant to your point and my sec­ond thought, which is that you are right, many things we might call “new age spiritu­al­ity” have the ex­act same jar­gon-coin­ing pat­tern in their writ­ing as ra­tio­nal­ist writ­ing does, with nearly ever au­thor striv­ing to ele­vate some metaphor to the level of word so that it can be­comes a part of a wider shared ap­proach to on­tol­ogy.

  This ac­tu­ally seems to sug­gest then that my story is too spe­cific and point­ing to Eliezer’s ten­dency to do this as a cause is maybe un­fair: it may be a ten­dency that ex­ists within many peo­ple, and there is some­thing similar about the kind of peo­ple or the so­cial in­cen­tives that are similar be­tween ra­tio­nal­ists and new age spiritu­al­ists that pro­duces this be­hav­ior.

• G Gordon Worley III 16 Apr 2019 16:27 UTC

  14 points

  on: Slack Club

  This archetype is eas­ily dis­tractible and does not co­op­er­ate with other in­stances of it­self, so an en­tire com­mu­nity of peo­ple con­form­ing to this archetype de­volves into valu­ing ab­strac­tion and spe­cial­ized jar­gon over solv­ing prob­lems.

  Ob­vi­ously there are ex­cep­tions to this, but as a first pass this seems pretty rea­son­able. For ex­am­ple, one thing I feel is go­ing on with a lot of posts on LessWrong and posts in the ra­tio­nal­ist di­as­pora is an at­tempt to write things the way Eliezer wrote them, speci­fi­cally with a mind to cre­at­ing new jar­gon to tag con­cepts.

  My sus­pi­cion is that peo­ple see that Eliezer gained a lot of pres­tige via his writ­ing, this is one of the things he does in his writ­ing (name con­cepts with un­usual names), and I sus­pect peo­ple make the (rea­son­able) as­sump­tion that if they do some­thing similar maybe they will gain pres­tige from their writ­ing tar­geted to other ra­tio­nal­ists.

  I don’t have a lot of ev­i­dence to back this up, other than to say I’ve caught my­self hav­ing the same temp­ta­tion at times, and I’ve thought a bit about this com­mon pat­tern I see in ra­tio­nal­ist writ­ing and tried to for­mu­late a the­ory of why it hap­pens that ac­counts not only for why we see it here but also why I don’t see it as much in other writ­ing com­mu­ni­ties.

• G Gordon Worley III 15 Apr 2019 21:54 UTC

  2 points

  in reply to: sirjackholland's comment on: High­lights from “In­te­gral Spiritu­al­ity”

  I don’t. In­te­gral Spiritu­al­ity might have some of what you’re look­ing for, but only in­ci­den­tally, since it’s re­ally try­ing to do some­thing else.

• G Gordon Worley III 14 Apr 2019 16:59 UTC

  6 points

  on: The Mean­ing(s) of Life

  Over­all it seems like you’re mak­ing a co­her­ent enough point (mean­ing­ness is sub­jec­tive), but I think the writ­ing style makes that a bit hard to pick out. I can tell you’ve got­ten some down­votes, and my guess is that it’s be­cause it’s hard read­ing this to tell much about why you think this is or re­ally given much in the way of spe­cific ar­gu­ments a per­son might en­gage with to­wards this point. I’m not say­ing I agree or dis­agree with you here, merely that I find it hard to fol­low your rea­son­ing be­cause it feels like there are many gaps in it in your writ­ing, even if there aren’t in your head.

  I see you are rather new to the site, so I say this be­cause I want to make sure you don’t end up bounc­ing off be­cause you get some down­votes. Peo­ple gen­er­ally re­spond well here to writ­ing that is proof-like: well struc­tured, aims to show some­thing, and is clear about its start­ing as­sump­tions (in­clud­ing as­sump­tions about the au­di­ence; I felt like this post was ar­gu­ing a point against an in­visi­ble other view that was never made very clear to me).

  Hope that helps.

High­lights from “In­te­gral Spiritu­al­ity”

• G Gordon Worley III 12 Apr 2019 16:02 UTC

  2 points

  in reply to: Chris_Leong's comment on: Agent Foun­da­tion Foun­da­tions and the Rocket Align­ment Problem

  “For­mally Stat­ing the AI Align­ment Prob­lem” is prob­a­bly the nicest in­tro­duc­tion, but if you want a preprint of a more for­mal ap­proach to how I think this mat­ters (with a cou­ple spe­cific cases), you might like this preprint (though note I am work­ing on get­ting this through to pub­li­ca­tion, have it halfway through re­view with a jour­nal, and al­though I’ve been time con­strained to make the re­view­ers’ sug­gested changes, I sus­pect the fi­nal ver­sion of this pa­per will be more like what you are look­ing for).

• G Gordon Worley III 10 Apr 2019 2:32 UTC

  2 points

  in reply to: Chris_Leong's comment on: Agent Foun­da­tion Foun­da­tions and the Rocket Align­ment Problem

  Both, al­though I mostly con­sider the former ques­tion set­tled (via a form of panpsy­chism that I point at in this post) and the lat­ter less about the tech­ni­cal de­tails of how AI could work and more about the philo­soph­i­cal pre­dic­tions of what will likely be true of AI (mostly be­cause it would be true of all com­plex, con­scious things).

  Also the “phe­nomenolog­i­cal” in the name sounded bet­ter to me than, say, “philo­soph­i­cal” or “con­ti­nen­tal” or some­thing else, so don’t get too hung up on it: it’s mostly a marker to say some­thing like “do­ing AI philos­o­phy from a place that much re­sem­bles the philos­o­phy of the folks who founded mod­ern phe­nomenol­ogy”, i.e. my philo­soph­i­cal lineage is more Kierkegaard, Hegel, Schopen­hauer, Husserl, and Sartre than Hume, White­head, Rus­sel, and Wittgen­stein.

• G Gordon Worley III 9 Apr 2019 21:55 UTC

  6 points

  on: Agent Foun­da­tion Foun­da­tions and the Rocket Align­ment Problem

  I’ve been a bit busy with other things lately, but this is ex­actly the kind of thing I’m try­ing to do.

• G Gordon Worley III 8 Apr 2019 19:02 UTC

  3 points

  on: Value Learn­ing is only Asymp­tot­i­cally Safe

  I sus­pect there are many more sources of risk that re­sult in only be­ing able to ap­proach com­plete safety than cos­mic rays, but this seems a rea­son­able ar­gu­ment for at least es­tab­lish­ing that the limit ex­ists so even if we dis­agree over whether some­thing more eas­ily con­trol­led by AI de­sign is a source of risk we don’t get con­fused and think that if we elimi­nate all risk from the de­sign that we sud­denly get perfect safety.

• G Gordon Worley III 4 Apr 2019 20:15 UTC

  7 points

  in reply to: shminux's comment on: A sig­nifi­cant idea

  This is a very out­side view on these ideas. I think from the in­side there’s a lot that of­ten sep­a­rates ob­vi­ously bo­gus ideas from pos­si­bly real ones. Ideas that might pan out are gen­er­ally plau­si­ble now given the ev­i­dence available, even if they can­not be proved, whereas bo­gus, crank ideas gen­er­ally ig­nore what we know to claim some­thing con­tra­dic­tory. This can get a bit tricky be­cause ideas of what peo­ple con­sider “known” can be a lit­tle fluid, but the dis­tinc­tion I’m try­ing to draw here is be­tween ideas that may con­tra­dict ex­ist­ing mod­els but agree with what we ob­serve and ideas that dis­agree with what we ob­serve (re­gard­less of whether they con­tra­dict ex­ist­ing mod­els), the former be­ing plau­si­ble, ahead-of-their-time ideas that might later be proven true, and the lat­ter be­ing clearly bo­gus.

  (Of course some­times, as in the case of not ob­serv­ing star par­al­lax with­out suffi­ciently pow­er­ful in­stru­ments, even our ob­ser­va­tions are a limit­ing fac­tor, but this does at least al­low us to make spe­cific pre­dic­tions that we should ex­pect to see some­thing if we had more pow­er­ful in­stru­ments, and would lead us to con­clude against a promis­ing idea if we got re­ally good ob­ser­va­tions that gen­er­ated dis­qual­ify­ing ev­i­dence.)

• G Gordon Worley III 4 Apr 2019 20:01 UTC

  2 points

  in reply to: John_Maxwell_IV's comment on: Co­her­ence ar­gu­ments do not im­ply goal-di­rected behavior

  What I like about this thread, and why I’m wor­ried about peo­ple read­ing this post and up­dat­ing away from think­ing that suffi­ciently pow­er­ful pro­cesses that don’t look like what we think are dan­ger­ous is safe, is that it helps make clear that Ro­hin seems to be mak­ing an ar­gu­ment that hinges on leaky or even con­fused ab­strac­tions. I’m not sure any of the rest of us have much bet­ter ab­strac­tions to offer that aren’t leaky, and I want to en­courage what Ro­hin does in this post of think­ing through the im­pli­ca­tions of the ab­strac­tions he’s us­ing to draw con­clu­sions that are spe­cific enough to be cri­tiqued, be­cause through a pro­cess like this we can get a clearer idea of where we have shared con­fu­sion and then work to re­solve it.

• G Gordon Worley III 4 Apr 2019 18:05 UTC

  3 points

  on: Model­ling Model Comparisons

  Type 1 seems to be de­scribing what I’d call a “struc­ture” which is an­other way of talk­ing about a pat­tern but in a cer­tain ab­stract sense. For ex­am­ple, con­sider the clas­sic math­e­mat­i­cian joke about topol­o­gists not be­ing able to dis­t­in­guish a donut from a coffee cup be­cause they have the same topolog­i­cal genus (at least, ideal­ized donuts and coffee cups do), genus-1.

  Type 2 seems to be de­scribing what I’d call a “sys­tem”, i.e. mul­ti­ple ob­jects in re­la­tion with each other com­ing to­gether to form a new ob­ject at a differ­ent level of ab­strac­tion.

  Although my think­ing has cer­tainly evolved a lot since then, I wrote about an is­sue that re­quired ad­dress­ing this topic a cou­ple years go, so you might find that in­ter­est­ing even if you’re not so in­ter­ested in the topic I was ad­dress­ing di­rectly.

• G Gordon Worley III 4 Apr 2019 17:44 UTC

  4 points

  on: A sig­nifi­cant idea

  I sus­pect panpsy­chism is in this boat. We have lots of philo­soph­i­cal rea­sons to think it makes sense, but mak­ing it the con­sen­sus re­quires over­com­ing two re­lated difficul­ties:

  • con­vinc­ing our­selves that “con­scious­ness” is less spe­cial and mag­i­cal than we cur­rently think it is

  • re­duc­ing con­scious­ness to some­thing eas­ily observable

  Part of the prob­lem seems to be we don’t have the abil­ity to ad­e­quately in­spect the most com­plex con­scious sys­tems, and un­til we do it will re­main pos­si­ble to keep claiming “yeah, but real con­scious­ness is spe­cial and not ev­ery­thing has it” be­cause we imag­ine the sim­ple pat­tern that strong the­o­ries of panpsy­chism pro­pose ex­plains con­scious­ness is in­suffi­cient to ex­plain the spe­cial­ness of hu­mans, an­i­mals, etc.

  (This is not to be con­fused with weak the­o­ries of panpsy­chism, which are woo and rea­son­ably dis­missed (based on cur­rent ev­i­dence) be­cause they pro­pose the ex­is­tence of phe­nom­ena we have not ob­served, like plants, rocks, and sys­tems be­ing as agen­tic as an­i­mals, but you know, in se­cret, or only on an­other plane of ex­is­tence.)

• G Gordon Worley III 4 Apr 2019 0:33 UTC

  5 points

  in reply to: Kaj_Sotala's comment on: De­grees of Freedom

  this feels like it’s mov­ing things to­wards op­ti­miza­tion and in­differ­ence.

  I came here to say some­thing like “I feel like this post sets up a false di­chotomy” and I think you’ve done a bet­ter job than I would have at ex­pli­cat­ing why it feels to me op­ti­miza­tion and in­differ­ence go to­gether and are not re­ally in op­po­si­tion, ex­cept from within the pris­ons of our own minds think­ing that they are in op­po­si­tion.

• G Gordon Worley III 2 Apr 2019 18:26 UTC

  2 points

  in reply to: An1lam's comment on: What are effec­tive strate­gies for miti­gat­ing the im­pact of acute sleep de­pri­va­tion on cog­ni­tion?

  I don’t have any liter­a­ture on it, but it has had that effect on me. That is, as long as I’m med­i­tat­ing reg­u­larly (I av­er­age about 45 min­utes a day of “se­ri­ous” med­i­ta­tion, and an­other 60 min­utes or so of “ca­sual” med­i­ta­tion) I find if I don’t get a full 10 hours I still of­ten won’t have sleep at­tacks (in fact I now nor­mally only sleep about 8 hours most nights) and I can sleep as lit­tle as 6 hours and still func­tion mostly nor­mally (but not do­ing that re­peat­edly, and I will al­most cer­tain have a sleep at­tack on those days).