--- Log opened Wed Jan 11 00:00:11 2023
00:26 < fenn> https://foresight.org/tech-tree/ these tech trees certainly contain a lot of information, but it's really hard just to look at them because of the interface
00:38 < fenn> hmph. the space tech tree doesn't have anything about tethers
00:46 < fenn> ok it's kinda useless. there isn't a clear set of "if a and b then c or d" style connections, it's mostly a jumble of related things connected by category nodes
00:47 < fenn> like you can't say "this technology must be developed first in order to get that capability" based on the data in the tree (mind map really)
00:48 < fenn> there are practically no nodes for existing technologies or capabilities
00:49 < fenn> i was expecting something like the rockwell integrated space plan
00:56 < fenn> not being able to collapse larger boxes down into nodes means you can't organize the tree on multiple levels
00:58 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has joined #hplusroadmap
01:04 < fenn> somewhere on the longevity tech tree i would want to see: [define aging] -> [understand aging mechanism] -> [interventions]
01:26 -!- Malvolio is now known as MLTVNZLV
01:26 -!- MLTVNZLV is now known as UUUV
01:28 -!- UUUV is now known as Malvolio
01:32 < fenn> the prototype longevity tech tree seems to have been much better laid out: https://miro.medium.com/max/1080/0*swi_IhFE-3n7sfck
01:41 < fenn> who is knowledgeable about human genetics, the nitty gritty details of sexual reproduction on a cellular level, or the production of gametes or embryos from somatic cells, and would be willing to answer a bunch of possibly dumb question?
01:42 < fenn> s
03:23 -!- FourFire [~FourFire@2a01:799:109e:ad00:aaa1:59ff:fe03:8b23] has joined #hplusroadmap
04:49 < L29Ah> Muaddib: are you knowledgeable about human genetics, the nitty gritty details of sexual reproduction on a cellular level, or the production of gametes or embryos from somatic cells, and would be willing to answer a bunch of possibly dumb question?
04:49 < Muaddib> L29Ah: Sure, why not? I'm always up for a good challenge.
04:49 < kanzure> fenn: yashgaroth
05:41 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0::4249] has joined #hplusroadmap
05:52 -!- FourFire [~FourFire@2a01:799:109e:ad00:aaa1:59ff:fe03:8b23] has quit [Quit: Client closed]
06:11 < muurkha> fenn: maybe the Foresight people need to play more video games
06:17 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has quit [Ping timeout: 260 seconds]
06:51 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has joined #hplusroadmap
08:14 -!- FourFire [~FourFire@2a01:799:109e:ad00:aaa1:59ff:fe03:8b23] has joined #hplusroadmap
09:03 -!- mrdata [~mrdata@user/mrdata] has quit [Ping timeout: 260 seconds]
09:03 -!- mrdata_ [~mrdata@135-23-182-185.cpe.pppoe.ca] has joined #hplusroadmap
09:05 -!- mrdata_ [~mrdata@135-23-182-185.cpe.pppoe.ca] has quit [Changing host]
09:05 -!- mrdata_ [~mrdata@user/mrdata] has joined #hplusroadmap
09:07 -!- mrdata_ is now known as mrdata
09:27 < kanzure> hmm someone is very adamantly telling me "it's all just transformer models and we're at human-level in the next 3 years"
09:28 < kanzure> absolutely convinced that we're done and will have AGI across all tasks and mediums
09:28 < kanzure> (he is, i mean, absolutely convinced)
09:29 < kanzure> the sentiment in the air is really interesting; i asked to be convinced and he goes "well nobody has written it down"
09:32 < kanzure> "it's just model compression and model distillation for transformers" etc
09:35 < L29Ah> kanzure: so make them bid in bitcoin in favor of that outcome
09:35 < kanzure> hm?
09:36 < L29Ah> if they're absolutely convinced, they can bid an arbitrarily large sum against your small sum of money to win it
09:38 < L29Ah> *bet
09:40 < kanzure> i think just buying/shorting nvidia would be easier for that
09:41 < L29Ah> maybe easier for you, but certainly less profitable
09:42 < L29Ah> (also ML ASICs are increasingly a thing, and Nvidia isn't leading there)
09:45 < kanzure> and that big causal inference chains are the only missing component for current transformer models to reach human-level performance across general tasks
09:48 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has quit [Ping timeout: 272 seconds]
10:08 -!- cthlolo [~lorogue@77.33.23.154.dhcp.fibianet.dk] has joined #hplusroadmap
10:21 -!- FourFire [~FourFire@2a01:799:109e:ad00:aaa1:59ff:fe03:8b23] has quit [Quit: Client closed]
10:30 < kanzure> https://news.ycombinator.com/item?id=31415478
11:22 < kanzure> .title
11:22 < saxo> Gato – A Generalist Agent | Hacker News
11:27 < kanzure> AI hardware companies right now should just raise money and then sit on it and do absolutely nothing- the benefits of waiting an extra year or two before implementing an algorithm in hardware is going to be pretty big, compared to doing an ASIC tapeout of the wrong thing entirely.
11:58 < kanzure> https://www.youtube.com/watch?v=c7ScUDYSRYo
11:58 < Muaddib> [c7ScUDYSRYo] Expert AI as a Healthcare Superpower (55:24)
12:01 < kanzure> https://www.lesswrong.com/posts/SkcM4hwgH3AP6iqjs/can-you-get-agi-from-a-transformer
12:02 < kanzure> "However, if you have a Transformer that more-or-less simulates the first 100 (or whatever) milliseconds of the neocortex's generative-model-search process, then that's all you can ever get." strange argument to make- i don't think anyone is claiming it simulates neocortex?
12:03 < kanzure> "I think there's an open question about the extent to which the algorithms underlying human intelligence in particular, and/or AGI more generally, can be built from operations similar to matrix multiplication (and a couple other operations). I'm kinda saying "no, it probably can't" while the scaling-is-all-you-need DNN enthusiasts are kinda saying "yes, it probably can"."
12:07 < kanzure> apparently this was the AGI definition from FTX Future Fund: "For any human who can do any job, there is a computer program (not necessarily the same one every time) that can do the same job for $25/hr or less."
12:10 < kanzure> https://www.lesswrong.com/posts/9Y4hFjjd22AH4Zm4q/agi-in-our-lifetimes-is-wishful-thinking
12:25 < kanzure> "This is what should have been told to Moravec:  "Sorry, even if your biology is correct, the assumption that future people can put in X amount of compute and get out Y result is not something you really know."  And that point did in fact just completely trash his ability to predict and time the future."
12:40 -!- cthlolo [~lorogue@77.33.23.154.dhcp.fibianet.dk] has quit [Read error: Connection reset by peer]
14:57 -!- TMM_ [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
14:57 -!- TMM_ [hp@amanda.tmm.cx] has joined #hplusroadmap
15:33 < fenn> was moravec in the business of making kurzweil-style trend extrapolations and timelines?
15:41 < muurkha> I think there are a lot of cases where you can calculate that future people can put in X amount of compute and get Y result
15:41 < muurkha> like, solving a large linear algebra problem with successive over-relaxation
15:42 < muurkha> maybe we'll have better algorithms in the future for large dense linear algebra problems, and we already have better ones for large sparse ones, but we almost certainly won't have to use worse ones
15:43 < muurkha> similarly if you have an algorithm that's running 10× slower than real time but producing the right results, you can confidently predict that if you could run it on a computer that was 10× as fast it could run in real time
15:45 < muurkha> you might also be able to do it in easier ways, like running it on 10 or 100 computers, or on a computer with different performance tradeoffs, or by optimizing the algorithm, but those are harder to predict
17:06 -!- TMM_ [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
17:07 -!- TMM_ [hp@amanda.tmm.cx] has joined #hplusroadmap
17:13 < kanzure> https://worldspiritsockpuppet.substack.com/p/we-dont-trade-with-ants
17:25 < kanzure> e11 job postings https://jobs.lever.co/convergentresearch
17:57 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0::4249] has quit [Quit: Leaving]
19:21 < kanzure> https://ai.stackexchange.com/questions/22877/how-much-computing-power-does-it-cost-to-run-gpt-3 https://www.reddit.com/r/MachineLearning/comments/gzb5uv/d_what_would_it_take_to_run_openais_gpt3_on/
19:22 < kanzure> about ~20 16 GB GPUs
19:28 < kanzure> "you could never fit ChatGPT on a single GPU.  You would need 5 80Gb A100 GPUs just to load the model and text."
20:00 < fenn> "In 2015, Yinong Yang successfully deactivated 16 specific genes in the white button mushroom to make them non-browning. Since he had not added any foreign-species (transgenic) DNA to his organism, the mushroom could not be regulated by the USDA under Section 340.2." apparently USDA only has jurisdiction to regulate "plant pests"
20:30 < fenn> does this have a name yet? it seems super important but i find myself stumbling over what to call it: take a pluripotent stem cell and generate an oocyte from it
20:30 < fenn> with induced pluripotency this enables a full lifecycle from somatic cells such as skin samples
20:31 < fenn> its been 6 years since this was demonstrated in mice, with healthy fertile mouse pups, and nobody's talking about it???
20:33 < kanzure> in vitro oogenesis
20:33 < fenn> thanks
20:33 < kanzure> i condemn you to reading and memorizing https://diyhpl.us/~bryan/papers2/bio/in-vitro-fertilization/
20:34 < kanzure> there is one document missing here from someone, struggling to remember the name
20:34 < kanzure> "will in boston"?
20:35 < kanzure> ah, https://docs.google.com/document/d/e/2PACX-1vT1hgf95phUa4e1nqf2VTfjFmUeSBwb7mihE86cU6bE1EI-EZteWm6UmeC18PJSs3Laog4rP7qKtT1j/pub
20:35 < kanzure> "in vitro gametogenesis" is the broader term.
20:35 < fenn> another process uses embryonic stem cells to create "sEmbryos" which as far as i can tell are just clone embryos - has this process been verified to create healthy fertile mouse pups?
20:35 < fenn> and can it also work with iPSCs?
20:36 < kanzure> ah damn that file was already in the directory :(
20:36 < fenn> that's a lotta papers
20:37 < kanzure> well there was this https://diyhpl.us/~bryan/papers2/bio/in-vitro-fertilization/VelociMouse:%20Fully%20embryonic%20stem%20cell-derived%20F0-generation%20mice%20obtained%20from%20the%20injection%20of%20embryonic%20stem%20cells%20into%20eight-cell-stage%20embryos%20-%202009.pdf
20:37 < fenn> your notes could be on the wiki (but then it would be a thing that people expect something of)
20:38 < fenn> oh ok i hadn't gotten to velocimouse yet
20:40 < kanzure> also this one seems relevant to your question about pluripotency into gametes,
20:40 < kanzure> "Haploidy in somatic cells is induced by mature oocytes in mice" https://www.nature.com/articles/s42003-022-03040-5
20:40 < kanzure> which also claims live mice pups
20:41 < kanzure> good night
20:41 < fenn> o/
20:41 < kanzure> o7
21:50 -!- stipa_ [~stipa@user/stipa] has joined #hplusroadmap
21:52 -!- stipa [~stipa@user/stipa] has quit [Ping timeout: 260 seconds]
21:52 -!- stipa_ is now known as stipa
22:01 < fenn> https://www.dermaclip.com/ prescription-only zip ties for skin, because we wouldn't want people to close their gaping wounds without permission
--- Log closed Thu Jan 12 00:00:12 2023