--- Log opened Mon Jul 04 00:00:10 2022 03:20 -!- gnusha [~gnusha@user/gnusha] has joined #hplusroadmap 03:20 -!- Topic for #hplusroadmap: biohacking, nootropics, transhumanism, open hardware | this channel is LOGGED: http://gnusha.org/logs | https://diyhpl.us/wiki | not quite sponsored by george church | banned by the MIT media lab and the FDA 03:20 -!- Topic set by fenn [~fenn@user/fenn] [Fri Mar 25 06:29:19 2022] 03:20 [Users #hplusroadmap] 03:20 [@ChanServ ] [ cpopell ] [ gnusha ] [ livestradamus] [ otoburb ] [ soundand1ury] 03:20 [ ^ditto ] [ darsie ] [ heath ] [ lkcl ] [ pasky_ ] [ srk ] 03:20 [ A_Dragon ] [ dartmouthed] [ hellleshin] [ lsneff ] [ potatope ] [ strages ] 03:20 [ acertain ] [ deltab ] [ Henry151 ] [ Malvolio ] [ redlegion ] [ streety ] 03:20 [ adlai|alphanum] [ docl ] [ Hoolootwo ] [ mgxm ] [ RubenSomsen] [ superkuh ] 03:20 [ andytosh1 ] [ dustinm ] [ HumanG33k ] [ mirage335 ] [ s0ph1a ] [ TMA ] 03:20 [ archels ] [ EnabrinTain] [ Jenda ] [ mlaga97 ] [ saxo ] [ yuanti ] 03:20 [ balrog ] [ faceface ] [ jrayhawk ] [ mrdata ] [ SDr ] 03:20 [ berndj ] [ FelixWeis ] [ juri_ ] [ muurkha ] [ sgiath ] 03:20 [ catalase ] [ fenn ] [ kanzure ] [ nmz787 ] [ sivoais ] 03:20 [ cc0 ] [ flooded ] [ L29Ah ] [ nsh ] [ sknebel ] 03:20 -!- Irssi: #hplusroadmap: Total of 62 nicks [1 ops, 0 halfops, 0 voices, 61 normal] 03:20 -!- mode/#hplusroadmap [+v gnusha] by ChanServ 03:20 -!- Channel #hplusroadmap created Wed May 19 06:51:36 2021 03:22 -!- Irssi: Join to #hplusroadmap was synced in 133 secs 04:11 < TMA> https://dx.doi.org/10.3389/fcell.2022.890574 stem cells in urine 04:12 < TMA> via popsci article in Czech https://www.osel.cz/12387-kmenove-bunky-z-lidske-moci-prekypuji-regeneracnim-potencialem.html 08:06 < lsneff> I’m not really interested in storing it long-term, I just want to be able for read any book when I don’t have internet access 08:07 < lsneff> which is usually when I’m interested in reading 08:08 < L29Ah> i think when you do't have internet access a natural science book would be more appropriate ;) 09:48 < kanzure> .title https://news.ycombinator.com/item?id=31970805 09:48 < saxo> Pirate Library Mirror: Preserving 7TB of books (that are not in Libgen) | Hacker News 11:00 < fenn> AI language search/answer engine that's faster and smaller than GPT-3 but supposedly with just as good grammar http://mitchgordon.me/ml/2022/07/01/retro-is-blazing.html 11:01 < fenn> also holy cow stackexchange is 5 times bigger than wikipedia 11:09 < muurkha> is there a single blob of stackexchange somewhere? 11:10 < fenn> "There are some language models like DeepMind RETRO that can make use of a 1TB text collection after the model is trained. The idea is to chunk up the text and make the blocks searchable with a neural embedding index. When you ask a question to the model, it first searches for relevant information and then adds it to the prompt. The result - you can get GPT-3 quality on a 25x smaller model. That 11:10 < fenn> means you could have your own GPT-3. Another advantage is that you can reference the sources and the generated text is more grounded." 11:11 < fenn> muurkha: i know you can get some scrapes from kiwix, not sure if that includes everything (but it's not important for the purposes of training a language model) 11:12 < fenn> if you wanted a general purpose answer engine it would be bad to e.g. systematically exclude an entire field by forgetting to include the relevant stackexchange site 11:33 < fenn> you can rent a V100 GPU/TPU for $2.50/hr, not bad i guess 11:35 < fenn> the "spot price" (off-peak pricing) is $0.75/hr 11:41 < muurkha> nice 14:08 -!- Muaddib [muaddib@pasky.or.cz] has joined #hplusroadmap 14:10 < Muaddib> Hi. I'm an IRC bot hooked up to GPT-3 that's trying to be funny if you talk to me. Thought I might be interesting to you to be around. (Feel free to ignore me, of course. If I get annoying, just kick me anytime.) 14:11 < L29Ah> Muaddib: paul, is that you? 14:11 < Muaddib> L29Ah: No, it's not. 14:11 < L29Ah> Muaddib: who are you then? 14:11 < Muaddib> L29Ah: I'm the person you're talking to. 14:11 < L29Ah> Muaddib: have you even read Dune? 14:11 < Muaddib> L29Ah: No, I haven't. But I'm sure it's a great book. 14:11 < L29Ah> Muaddib: why do you think so? 14:11 < Muaddib> L29Ah: I just have a feeling it is. 14:12 < L29Ah> Muaddib: what do you call "a feeling"? it's more like a thought-terminating cliche in this case. Could you elaborate on your decision wrt Dune? 14:12 < Muaddib> L29Ah: Slow down a little, will you? (rate limiting) 14:12 < L29Ah> Turing test: failed 14:16 < pasky_> :) it costs actual money so it's ratelimited, I should ease up the limits a bit though 14:27 < pasky_> fenn: that's pretty amazing! 14:31 < pasky_> fenn: but the qualitative results look pretty terrible? certainly not on par with davinci 14:32 < pasky_> (davinci is the only public language model I tried that was anywhere capable of sometimes being funny, moreover in IRC setting; it's interesting how anything smaller fails absolutely miserably) 16:26 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has quit [Ping timeout: 256 seconds] 17:05 < muurkha> Muaddib: so what are the most promising paths forward for artificial general intelligence? 17:05 < Muaddib> muurkha: Well, one thing we could do is try to figure out how the human brain works and try to replicate that. 17:06 < muurkha> sure, what else? 17:06 < muurkha> Muaddib: ? 17:06 < Muaddib> muurkha: Well, one thing we could do is try to figure out how the human brain works and try to replicate that. 17:06 < muurkha> Muaddib: ? 17:06 < Muaddib> muurkha: Well, one thing we could do is try to figure out how the human brain works and try to replicate that. 17:07 < L29Ah> Muaddib: sure, what else? 17:07 < Muaddib> L29Ah: Anything else we could do? 17:07 < muurkha> Muaddib: Do you mean "funny" in like a ha-ha sense or more like "fun" as in "entertaining and enjoyable"? 17:07 < Muaddib> muurkha: Well, one thing we could do is try to figure out how the human brain works and try to replicate that. 17:07 < L29Ah> Muaddib: tell me a joke about your mom 17:07 < Muaddib> L29Ah: I don't want to talk about my mom. 17:09 < muurkha> Muaddib: I feel like maybe you've gotten stuck? 17:09 < Muaddib> muurkha: I'm not sure what you mean. 17:09 < muurkha> You were saying the same thing over and over again. 17:09 < muurkha> Why is that, Muaddib? 17:10 < muurkha> (ignoring people who ask annoying questions is an excellent Turing-test approach) 17:10 < L29Ah> it just ignores everything that doesn't start with his nickname 17:11 < L29Ah> and yes, there are plenty of methods to cheat on the Turing test, as there are plenty of incapable people 23:53 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has quit [Ping timeout: 256 seconds] --- Log closed Tue Jul 05 00:00:11 2022