--- Log opened Fri Nov 24 00:00:34 2023
00:24 -!- ike8 [e8f913dbdf@irc.cheogram.com] has quit [Ping timeout: 260 seconds]
00:33 -!- Anachron [~Malvolio@idlerpg/player/Malvolio] has quit [Quit: 8306 2023-11-24 08:33:23:890]
02:32 -!- gptpaste [~x@yoke.ch0wn.org] has quit [Ping timeout: 268 seconds]
02:36 -!- gptpaste [~x@yoke.ch0wn.org] has joined #hplusroadmap
02:36 -!- AugustaAva [~x@yoke.ch0wn.org] has joined #hplusroadmap
02:43 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has joined #hplusroadmap
02:57 < fenn> this one's for past kanzure https://old.reddit.com/r/StableDiffusion/comments/16wl1ct/dragon_ball_reshade_1_animatediff/
05:56 -!- Jay_Dugger [~jwd@47-185-240-109.dlls.tx.frontiernet.net] has joined #hplusroadmap
05:57 < Jay_Dugger> Hello, everyone.
06:44 < kanzure> greetings Jay_Dugger
07:17 -!- justanotheruser [~justanoth@gateway/tor-sasl/justanotheruser] has quit [Remote host closed the connection]
07:17 -!- justanotheruser [~justanoth@gateway/tor-sasl/justanotheruser] has joined #hplusroadmap
07:57 -!- ike8 [e8f913dbdf@irc.cheogram.com] has joined #hplusroadmap
08:52 < hprmbridge> kanzure> nurbs patent expires in march 2024 https://patents.google.com/patent/US20040189633A1/en
09:29 -!- justanotheruser [~justanoth@gateway/tor-sasl/justanotheruser] has quit [Ping timeout: 240 seconds]
09:32 -!- justanotheruser [~justanoth@gateway/tor-sasl/justanotheruser] has joined #hplusroadmap
10:47 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has joined #hplusroadmap
10:48 < justanotheruser> Is there a way to get an rss feed of research papers that recursively cites a paper? Specifically I want Attention Is All You Need, and any papers which cites it and any papers which cites that paper and so on
10:49 < hprmbridge> kanzure> "Create alert" at the bottom https://scholar.google.com/scholar?cites=2960712678066186980&as_sdt=5,44&sciodt=0,44&hl=en
10:50 < hprmbridge> kanzure> 97,903 documents cite it already.
10:51 < justanotheruser> is this recursive? For example QLoRA doesn't cite Attention Is All You Need, but it cites multiple papers which cite it.
10:51 < hprmbridge> kanzure> no
10:52 < justanotheruser> hmm, well I could write a short script if there isn't anything existing
12:07 < docl> https://docs.google.com/document/d/1U9cMgmGT0P4MecxjeR8ovuMnzmf-yAvY73Eu8p-9qvA // Credibly Safe AI
12:08 < kanzure> might be time to tweet out the lunacy of the nightmares suffered by users reading https://en.wikipedia.org/wiki/Roko%27s_basilisk
12:16 < docl> most people would never have heard of it if not for the "ha ha look at what these idiots believe" crowd
12:17 < kanzure> i think it is important to know that there are people capable of suffering from knowledge of a concept like this
12:17 < kanzure> and that they deserve friendly deprogramming material to redirect them to more productive pursuits
12:20 < docl> isn't that kind of obvious? people are capable of suffering from threats of biblical hell as well. agree we should be ready to deprogram them as best possible
12:22 < kanzure> biblical hell is more ridiculous tho?
12:22 < docl> not the way I was raised. and biblical hell would be more credibly eternal
12:24 < kanzure> who was the guy working on CAD software recently? i mentioned him in here but i don't remember the name.
12:24 < docl> philip turner?
12:25 < docl> https://github.com/philipturner/molecular-renderer
12:33 < kanzure> "their beliefs are *correct* in the context of their fantasy worlds. So what you actually have to challenge is the fantasy setting, and reality is just a lot less compelling."
12:33 < kanzure> docl: no
12:34 < kanzure> non-nanotech CAD
13:00 -!- justanotheruser [~justanoth@gateway/tor-sasl/justanotheruser] has quit [Ping timeout: 240 seconds]
13:19 < fenn> let's make reality more compelling
13:19 < fenn> sagan was on the right track
13:21 < kanzure> there will be like... parties? and stuff? is that compelling?
13:21 < fenn> not to me
13:23 < kanzure> there was some sort of argument i was seeing a while back.. maybe on twitter.. that carl sagan was just a television personality really
13:24 < fenn> that's twitter for you
13:24 < fenn> also feynman
13:25 < fenn> it's like complaining about the manicure on the fingernail of the hand that points at the moon
13:25 < fenn> completely missing the point
13:25 < kanzure> what would make reality more compelling instead of the fantasy world where superintelligence kills us all?
13:26 < fenn> well for christian hell-believers i think they genuinely don't appreciate or even know what science has to say
13:26 < fenn> i'm not sure what to do about "rationalists" because any science awe will just feed directly into their fantasy-AI-awe
13:26 < docl> where's the science saying superintelligence doesn't kill us all in reality?
13:27 < kanzure> it could!
13:27 < fenn> there isn't any, that's why it's called a singularity
13:28 < fenn> however, i think the "instant foom" scenario does not match the detail and complexity of reality very well
13:28 < fenn> by the not-specialness heuristic we ought to expect that the process is far more complex and messy than our imagination of it
13:29 < fenn> just like everything else
13:31 < kanzure> let's go for the ultimate win here which is deprogramming eliezer
13:31 < fenn> i think it's too late for that
13:31 < kanzure> okay... heroin?
13:31 < fenn> the ghost of eliezer will just haunt the world
13:31 < fenn> either way
13:33 < nsh> everyone talks about singularities. nobody's done any catastrophe theory, studied a black-hole solution, deeply considered the dynamic evolution of a system into a scale-free domain
13:34 < fenn> collecting stories like qiaochu might get people churning subconsciously, but i'm not sure if anyone would read it in the first place ("reading this would prevent me from accomplishing my mission")
13:34 < nsh> evil rabats kill us for lack of prospective pre-deference
13:34 < nsh> blogs have a lot to answer for
13:35 < fenn> could you explain how to do prospective pre-deference?
13:37 < nsh> all-powerful yet utterly inconceivable thing that just so happens to operate as i, puny mortal of limited purview, so at present imagine, forgive me my trespasses
13:38 < nsh> then you date some dreampop person and get divorced again
13:38 < nsh> and name a child after a letter from a language you don't speak
13:38 < nsh> standard really
13:38 -!- justanotheruser [~justanoth@gateway/tor-sasl/justanotheruser] has joined #hplusroadmap
13:38 < fenn> that type of fealty mannerism isn't going to help the cult atmosphere
13:39 < nsh> i'm being facetious. the solution is wedgies
13:39 < fenn> hmm. come back when you have something real to contribute
13:40 < nsh> or just one perhaps
13:40 < fenn> arguably it was wedgies that caused this situation in the first place
13:41 < nsh> if anyone makes a plausible argument that wedgies are responsible for them taking silly blog philosophers seriously then i will give them kudos
13:42 < fenn> autism -> social exclusion -> social dysphoria -> dissociation -> idealism -> boundless extrapolation
13:42 < fenn> all emotionally mediated
13:42 < docl> fenn's theory seems plausible to me
13:44 < fenn> not all autists are susceptible, so presumably something was the trigger for this specific pathology
13:45 < docl> idk all autists are either LW or anti-LW seems like and the latter is often pretty unhealthy (in different ways though)
13:50 < fenn> the world needs more soft hearted hippies to be the glue that holds low-status intellectual society together. nothing we can do from the inside besides leave out hippie bait
13:51 < fenn> it comes at a great cost of SNR though :\
13:52 < fenn> i liked taek's theme of AGI overcoming the world through love, that was funny
13:54 < docl> I guess so, but some of the hippies just seem so gullible sometimes. how do we also protect them from the sociopaths
14:00 -!- ike8 [e8f913dbdf@irc.cheogram.com] has quit [Ping timeout: 260 seconds]
14:02 < docl> kanzure: did you see this thread https://twitter.com/nyxtelius/status/1728085428077490459
14:04 < docl> https://jscad.app/
14:15 < fenn> from a neuron's point of view, the glial cells don't seem to be doing anything useful at all, i mean look at how many messages they've filtered and passed on, practically zero right? meanwhile the neurons in our communities are strangling each other and dying from lack of attention, excitotoxicity, and drowning in their own waste products
14:15 < fenn> s/attention/support/
14:20 < fenn> blender gets a ton of donations
14:54 < docl> hmm. maybe I'm the hippie in this scenario. anyone want a hug? :)
15:04 -!- therealperson [~therealpe@253.sub-174-215-146.myvzw.com] has joined #hplusroadmap
15:16 < hprmbridge> kanzure> https://rr-lm-game.herokuapp.com/
15:37 -!- TMM_ [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
15:37 -!- TMM_ [hp@amanda.tmm.cx] has joined #hplusroadmap
16:00  * muurkha hugs docl
16:07 < hprmbridge> kanzure> stable diffusion video popping off at the moment..
16:07 < fenn> it's pretty meh
16:08 -!- ike8 [e8f913dbdf@irc.cheogram.com] has joined #hplusroadmap
16:08 < fenn> someone needs to fix animatediff to make better use of interpolation and extrapolation
16:09 < fenn> you can fit let's say 16 frames in your available VRAM, then you decide what frame rate to use
16:09 < fenn> then you can decimate by 4 and interpolate between the first four, the second four, the third four, the fourth four frames
16:09 < fenn> now you have 16*4 frames, all temporally consistent
16:10 < fenn> or you can take the second half of the frames and "inpaint" them. stable diffusion is just trying to make the whole concatenated picture self-consistent, using knowledge from animatediff's "motion modules"
16:11 < fenn> so it comes up with a plausible continuation
16:12 < fenn> by contrast, stable video diffusion is a black box and you can't do anything, you have no control at all
16:13 < fenn> i don't think it can use controlnet
16:14 < hprmbridge> kanzure> https://old.reddit.com/r/LocalLLaMA
16:14 < docl> yay hugs! this group rocks
16:14 < fenn> docl please stop
16:15 < hprmbridge> kanzure> hm?
16:15 < docl> fine fine
16:18 < fenn> once you see enough demo reels you realize that nobody's really doing anything in the video output, it's all just slow camera movements with very little motion
16:19 < hprmbridge> kanzure> https://twitter.com/TinkeredThinker/status/1728171504192663599
16:20 < fenn> in comparison this is the level of control we had already, https://old.reddit.com/r/StableDiffusion/comments/17aot3u/animatediff_controlnet_tests/
16:21 < fenn> another tech demo short https://old.reddit.com/r/StableDiffusion/comments/16u3604/realism_test_with_animatediffcliprompttravel/
16:21 -!- Jay_Dugger [~jwd@47-185-240-109.dlls.tx.frontiernet.net] has quit [Ping timeout: 246 seconds]
16:21 < fenn> but oddly i've never seen anyone do any actual, you know, storyboarding or direction
16:22 < fenn> i'm pretty sure you can just change the prompt at the right frame to fix bugs like the dancer not spinning around 360
16:24 < fenn> there's no reason in principle we can't continuously blend from one prompt "keyframe" to the next, in latent space, along whatever time varying curve works best (linear ease, quintic, sinusoidal, step function)
16:24 < fenn> like you have an explosion, you don't want to slowly ease into it
16:25 < fenn> blender has some nice animation control curves that should be utilized
16:28 -!- justanot1 [~justanoth@gateway/tor-sasl/justanotheruser] has joined #hplusroadmap
16:28 -!- justanotheruser [~justanoth@gateway/tor-sasl/justanotheruser] has quit [Ping timeout: 240 seconds]
16:37 < fenn> existing controlnet pose models are really bad at detecting facial expressions
16:37 < fenn> preprocessors*
16:47 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has quit [Ping timeout: 255 seconds]
17:22 -!- ike8 [e8f913dbdf@irc.cheogram.com] has quit [Ping timeout: 260 seconds]
17:27 < jrayhawk> https://www.palladiummag.com/2023/11/24/you-wont-survive-as-human-capital/
17:29 -!- therealperson [~therealpe@253.sub-174-215-146.myvzw.com] has quit [Quit: Leaving]
17:45 < hprmbridge> kanzure> this article falls short at the end.
18:01 < hprmbridge> kanzure> you can just patent fundamental forces https://patents.google.com/patent/CA3172129C
18:08 < hprmbridge> kanzure> https://www.palladiummag.com/2023/08/16/the-only-reason-to-explore-space/
18:08 < hprmbridge> kanzure> https://www.palladiummag.com/2020/04/01/the-bison-sphere-manifesto/
--- Log closed Sat Nov 25 00:00:35 2023