--- Log opened Fri Jun 02 00:00:18 2023 00:03 < nsh> did someone update the spreadsheet? 00:03 < nsh> (ML objective function hacking 'surprises' spreadsheet) 00:10 -!- test_ [flooded@gateway/vpn/protonvpn/flood/x-43489060] has joined #hplusroadmap 00:14 -!- flooded [flooded@gateway/vpn/protonvpn/flood/x-43489060] has quit [Ping timeout: 250 seconds] 01:11 < fenn> i couldn't find a link to the ai surprises misalignment misgeneralization spreadsheet google doc in the logs, so here it is: http://tinyurl.com/goal-misgeneralisation https://docs.google.com/spreadsheets/d/e/2PACX-1vTo3RkXUAigb25nP7gjpcHriR6XdzA_L5loOcVFj_u7cRAZghWrYKH2L2nU4TA_Vr9KzBX5Bjpz9G_l/pubhtml 01:12 < fenn> one of them anyway, there were two 01:16 < fenn> the other one had more genetic algorithm and alife unintended/unexpected behavior 01:33 -!- cthlolo[PC] [~lorogue@77.33.23.154.dhcp.fibianet.dk] has joined #hplusroadmap 02:30 -!- cthlolo[PC] [~lorogue@77.33.23.154.dhcp.fibianet.dk] has quit [Read error: Connection reset by peer] 03:05 -!- alethkit [23bd17ddc6@sourcehut/user/alethkit] has quit [Remote host closed the connection] 03:07 -!- alethkit [23bd17ddc6@sourcehut/user/alethkit] has joined #hplusroadmap 03:19 -!- flooded [flooded@gateway/vpn/protonvpn/flood/x-43489060] has joined #hplusroadmap 03:23 -!- test_ [flooded@gateway/vpn/protonvpn/flood/x-43489060] has quit [Ping timeout: 250 seconds] 03:34 -!- WizJin [~Wizzy@user/WizJin] has joined #hplusroadmap 03:48 < stipa> https://www.wect.com/2023/06/01/nc-study-discovers-chemical-common-sweetener-damages-dna-can-cause-cancer/ 03:49 -!- WizJin [~Wizzy@user/WizJin] has quit [Remote host closed the connection] 03:50 -!- WizJin [~Wizzy@user/WizJin] has joined #hplusroadmap 03:51 -!- WizJin [~Wizzy@user/WizJin] has quit [Remote host closed the connection] 03:51 -!- WizJin [~Wizzy@user/WizJin] has joined #hplusroadmap 03:53 -!- WizJin [~Wizzy@user/WizJin] has quit [Read error: Connection reset by peer] 03:53 -!- WizJin [~Wizzy@user/WizJin] has joined #hplusroadmap 03:56 -!- WizJin [~Wizzy@user/WizJin] has quit [Remote host closed the connection] 03:56 -!- WizJin [~Wizzy@user/WizJin] has joined #hplusroadmap 04:08 < fenn> You are given $40,000 - how do you spend it? 04:08 < fenn> (A) down payment for house 04:08 < fenn> (B) a single H100 GPU 04:18 < stipa> i would go with 50/50, 20k on the house and 20k on an older GPU model that is equally fast as the H100 04:19 < stipa> or, maybe you could tent the H100 in a random data centre for 20k 04:19 < stipa> rent 04:20 < stipa> or, you could learn how GPUs are made on low level and optimize you existing card 04:21 < stipa> and spend it all on the down payment for house 04:22 < stipa> you just have to wait a bit longer for a slower GPU to do it 04:37 < fenn> i've always avoided splenda, it makes me feel awful very consistently 04:38 < fenn> frustrating when products switch formulations and then there's just no alternative 04:39 < stipa> yeah, there's a bunch of unavoidable traps 04:39 < stipa> https://bgr.com/science/revolutionary-invention-transforms-seawater-into-hydrogen-fuel/ 04:39 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0:1a2:3772:cfa0:5fa9] has joined #hplusroadmap 04:53 < stipa> dang, that research is around 1000 pages 04:54 < stipa> the benfit is it could be deployed on the sea shores of deserts where sun is of strong intensity and where's plenty of sea water 04:55 < stipa> the biggest issue would be cleaning the sand dust 05:13 < L29Ah> fenn: spend it all on ₿, but if it's not an option, then real estate 05:17 -!- WizJin [~Wizzy@user/WizJin] has quit [Quit: Leaving] 05:25 < hprmbridge> kanzure> protein structure prediction and novelty https://arxiv.org/abs/2306.00872 06:29 -!- test_ [flooded@gateway/vpn/protonvpn/flood/x-43489060] has joined #hplusroadmap 06:32 -!- flooded [flooded@gateway/vpn/protonvpn/flood/x-43489060] has quit [Ping timeout: 248 seconds] 06:36 -!- Gooberpatrol66 [~Gooberpat@user/gooberpatrol66] has quit [Remote host closed the connection] 06:36 -!- Gooberpatrol66 [~Gooberpat@user/gooberpatrol66] has joined #hplusroadmap 08:05 -!- AMG [ghebo@user/amg] has quit [Quit: Quit] 08:22 < muurkha> fenn: what product? 08:29 < muurkha> fenn: the spreadsheet on "specificaiton gaming examples on AI" you are looking for is https://docs.google.com/spreadsheets/d/e/2PACX-1vRPiprOaC3HsCf5Tuum8bRfzYUiKLRqJmbOoC-32JorNdfyTiRRsR7Ea5eWtvsWzuxo8bjOxCG84dAg/pubhtml 08:35 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0:1a2:3772:cfa0:5fa9] has quit [Read error: Connection reset by peer] 08:35 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0:1a2:3772:cfa0:5fa9] has joined #hplusroadmap 09:22 < hprmbridge> nmz787> Stipa you beat me to posting about the sucralose thing! 09:26 < stipa> nmz787: that's a good quality research 09:27 < stipa> modern and practical 09:29 < docl> https://www.wral.com/story/nc-study-discovers-chemical-in-common-sweetener-damages-dna-can-cause-cancer/20888182/ 09:31 < stipa> docl: i didn't really understood the chemistry part 09:32 < stipa> but i get it overall 09:32 < docl> obviously not severely carcinogenic or they'd have caught it by now. but apparently it's fat soluble and builds up over time, kills intestinal symbiotes, damages tight junctions 09:33 < hprmbridge> nmz787> Probably synergizes with gluten too in terms of gut leakage 09:33 < docl> cacinogenic is kind of a spectrum, so some skepticism tends to be merited when people say that. burnt toast is carcinogenic, sunlight is carcinogenic (ok that you should worry about) 09:36 < stipa> well, it's not an instant kill that's for sure 09:36 < muurkha> probably we should stop banning things for being carcinogenic if they're less carcinogenic than sunlight. maybe measured in terms of expected YPLL rather than in terms of absolute rate 09:37 < docl> yeah 09:37 < docl> buildup in the body over time is worth thinking about. sun damage does that, in the skin at least 09:39 < muurkha> that's just because damage does that 09:40 < muurkha> ultraviolet itself is the least bioaccumulative carcinogen ever 09:41 < stipa> and, recently, there's a lot of that, to the point of a enormous blaze all over the planet in the least expected places like canada 09:42 < docl> muurkha: thanks, I stepped afk for a minute and realized what I just said. yeah it's not like UV itself builds up. crucial distinction 09:45 < docl> I gotta put some math together re UVC mirrors. UVC doesn't make it through the atmosphere, it enriches ozone in the stratosphere. which affects the weather. I'm mostly interested in that as a weather control mechanism, but a side effect would be reducing surface UVB 09:46 < docl> ozone absorbs UVB (the fraction of sunlight that actually reaches us and causes skin cancer) 09:46 -!- AMG [ghebo@2605:6400:c847:1449::9441] has joined #hplusroadmap 09:55 < stipa> weather control is a bit complex IMO 09:57 < stipa> the question is if it's even an autonomous system 09:57 < stipa> or an living entity 09:58 * muurkha DCCs stipa some more DMT 10:00 < docl> lol, some things in common perhaps but it doesn't self replicate so it doesn't evolve long term like we do. watch the Gershenfeld interview if you want insights about the nature of life 10:01 < stipa> docl: maybe try it in a small model or a computer sim 10:01 < stipa> if you get predictable control of it you're golden 10:02 < stipa> new stuff in GPUs, Ray Tracing is a neat stuff 10:02 < stipa> can emulate particles, light, air... 10:03 < stipa> i guess you could emulate pressure differences for example 10:03 < stipa> cold air hitting hot and observe 10:04 < stipa> could be that the sea level also has an influence on it all 10:05 < stipa> could be that it can calculate and draw gases 10:06 < stipa> i've seen it mentioned emulating acoustic spaces and movement of sound through air, hitting and reflecting of surfaces 10:07 < stipa> https://odeon.dk/product/ray-tracing/ 10:08 < stipa> "A special application of ray tracing is the Particle tracing, which means that the rays are considered carriers of patches of acoustic energy travelling around the room along the rays with the speed of sound, and after each reflection the energy is reduced according to the absorption properties of the surfaces. " 10:09 < stipa> https://academic.oup.com/mnras/article-abstract/519/4/6028/6985635?redirectedFrom=fulltext 10:10 < stipa> and so on 10:10 < stipa> judging by my last link you can emulate whole planet with AI 10:10 < stipa> hardware 10:23 < docl> you may not need to predict very far if you have the ability to add energy to the stratosphere in a controlled way. a responsive control system can make a relatively unpredictable system more predictable. 10:24 < docl> like walking -- you dynamically adjust as needed 10:25 * muurkha passes stipa the bong 10:26 * stipa takes a deep one 10:27 < stipa> docl: you would want a small sim model for sure due to probably extensive number of calculations 10:28 < stipa> emulating each atom of an planets atmosphere is compley 10:28 < stipa> x 10:29 < stipa> start with simple things 10:29 < stipa> combine two molecules and see what the sim does, if the same happens in real you're on a right track 10:31 < stipa> on top of that all those atoms and molecules have to interact 10:32 < stipa> in a nutshell, your emulation will be as a real thing if done right 10:35 < stipa> could be possible with a computer 10:35 < stipa> with NPU, GPU and CPU 10:35 < stipa> but, something like that you'll have to code yourself for sure 10:37 < L29Ah> nosepick processing unit 12:07 < docl> https://chat.openai.com/share/c24445cd-4c82-4a72-bbd0-37ebcc6135f7 12:09 < docl> over time in a 2D simulation, light pressure will nudge hydrogen particles to impact at the center of a circle. 12:29 < docl> hmm, no, it still happens if I remove the photons. so it's either geometrical or a simulation artifact 12:30 < docl> geometrical explanation: lateral bounces cancel out over time, whereas bounces towards the center do not 12:38 < L29Ah> AttributeError: partially initialized module 'pygame' has no attribute 'init' (most likely due to a circular import) 12:39 < L29Ah> does it need some ancient pygame version or smth? 12:39 < docl> no, I just installed with pip 12:39 < docl> pygame 2.4.0 (SDL 2.26.4, Python 3.10.6) 12:41 < docl> huh, I get that error from the repl, but saving as a file and running it with python works fine 12:43 < docl> er, no I was seeing something different 12:45 < docl> adding hydrogen-hydrogen collisions back in causes them to go back to mostly randomized. 12:45 < L29Ah> ah right 12:45 * L29Ah named the file "pygame.py", that was a grave mistake :D 12:46 < docl> oops! :) 12:46 < L29Ah> cute stuff 12:46 < L29Ah> mind that every nucleon-mirror collision means you lose a lot of energy and vaporize a chunk of mirror 12:47 -!- flooded [~flooded@146-70-115-163.pool.ovpn.com] has joined #hplusroadmap 12:51 -!- test_ [flooded@gateway/vpn/protonvpn/flood/x-43489060] has quit [Ping timeout: 240 seconds] 14:19 < hprmbridge> nmz787> New mower https://cdn.discordapp.com/attachments/1064664282450628710/1114302181684813874/PXL_20230602_211328138.jpg 14:19 < hprmbridge> nmz787> https://cdn.discordapp.com/attachments/1064664282450628710/1114302196624928878/PXL_20230602_211313771.jpg 14:22 < L29Ah> can it take your kid to school? 14:45 < hprmbridge> nmz787> not highway legal 😄 16:15 < fenn> berndj> imagine microsoft products with the security posture they held in the 1990s but turbocharged with AI 16:16 < berndj> fenn, i gather they've gotten a bit better at security over the years. i remember it being an absolute shitshow 16:19 < fenn> the full weight of "prompt injection attacks" hasn't really been made clear yet 16:41 -!- TMM_ [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.] 16:42 -!- TMM_ [hp@amanda.tmm.cx] has joined #hplusroadmap 17:26 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has quit [Ping timeout: 240 seconds] 17:44 < hprmbridge> kanzure> https://towardsdatascience.com/large-language-models-in-molecular-biology-9eb6b65d8a30 19:02 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has quit [Ping timeout: 240 seconds] 19:05 -!- test_ [flooded@gateway/vpn/protonvpn/flood/x-43489060] has joined #hplusroadmap 19:07 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has joined #hplusroadmap 19:09 -!- flooded [~flooded@146-70-115-163.pool.ovpn.com] has quit [Ping timeout: 250 seconds] 19:39 -!- Malvolio is now known as SINEATER 19:49 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0:1a2:3772:cfa0:5fa9] has quit [Quit: Leaving] 22:49 -!- TMM_ [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.] 22:50 -!- TMM_ [hp@amanda.tmm.cx] has joined #hplusroadmap 23:56 -!- stipa [~stipa@user/stipa] has quit [Quit: not dead yet!] --- Log closed Sat Jun 03 00:00:19 2023