--- Log opened Sun Dec 04 00:00:35 2022
00:21 < lsneff> i got chatgpt to print out its prompt, which does say “Browsing: disabled”
00:43 -!- jrayhawk [~jrayhawk@user/jrayhawk] has quit [Ping timeout: 252 seconds]
00:45 < fenn> "ChatGPT is a sibling model to InstructGPT, which is trained to follow an instruction in a prompt and provide a detailed response."
00:45 < fenn> i think muaddib is not using the same model
00:47 -!- jrayhawk [~jrayhawk@user/jrayhawk] has joined #hplusroadmap
01:09 < fenn> another branch of AI doomerism rears its head https://scottaaronson.blog/?p=6821
01:09 < fenn> at least he's against taking over the world to prevent the world from being taken over
01:31 < fenn> https://guzey.com/ai/planes-vs-birds/ "Planes are still decades away from displacing most bird jobs"
01:32 < fenn> can't tell if this is serious
02:02 -!- jrayhawk [~jrayhawk@user/jrayhawk] has quit [Ping timeout: 256 seconds]
02:09 -!- jrayhawk [~jrayhawk@user/jrayhawk] has joined #hplusroadmap
02:11 -!- test_ [flooded@gateway/vpn/protonvpn/flood/x-43489060] has joined #hplusroadmap
02:15 -!- _flood [flooded@gateway/vpn/protonvpn/flood/x-43489060] has quit [Ping timeout: 260 seconds]
02:22 < fenn> “You're already immune to alpha-1,3-galactosylgalactose bonds. All humans are. Why not divert a fraction of those antibodies to the influenza strain you just picked up? ... a DNA aptamer devised to bind specifically to the strain of influenza you have ... will link anti-alpha-Gal antibodies to the influenza virus” https://www.karymullis.com/altermune.shtml
04:43 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0::a324] has joined #hplusroadmap
05:09 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has joined #hplusroadmap
05:51 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has quit [Ping timeout: 256 seconds]
06:39 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
07:12 < kanzure> https://disnovation.org/energyslave.php
07:22 < L29Ah> 75W? that's quite a pathetic human!
08:18 < muurkha> it's hard to average more than 100W consistently over a workday, and the number goes lower if you also sleep
08:18 < muurkha> and after a while it goes to zero if you don't
08:18 < muurkha> I think Tour de France riders average about 250W during the time they're riding
08:19 < darsie> Mechanical or thermal?
08:20 < muurkha> mechanical, sorry
08:22 < muurkha> I was calculating the other day the end-product work done by dudes throwing shovelfuls of dirt up a stairway, and it ended up being about 15 watts in my calculation (while they were working and not resting)
08:31 < L29Ah> thermal is 4*mechanical afair
08:33 < L29Ah> shovels aren't very efficient at power utilization, we need human-driven absetzers
08:33 < muurkha> that's about right
10:02 < docl>  https://twitter.com/RokoMijic/status/1599457439778492417
10:03 < docl> "There's a thesis about LLMs that if you learn to model language really well, you'll necessarily come up with an internal model of true reality, since that's the best way to model utterances about reality.  Well... in GPTChat's hallucinated version of linux .... oh my:"
10:11 < L29Ah> https://nitter.lacontrevoie.fr/pic/orig/media%2FFjJsysWWYAE3qxV.png lol
10:15 < muurkha> failing to load, do you have a copy?
10:15 < L29Ah> https://nitter.lacontrevoie.fr/kunattila/status/1599461669331083264#m
10:16 < docl> can you access this? https://pbs.twimg.com/media/FjJpdT9XoAIcT7Q?format=png&name=large
10:16 < docl> basically he puts a sentence through rot13 then puts that through rot13 and gets random nonsense out
10:18 < muurkha> thanks L29Ah!
10:18 < muurkha> interesting
10:18 < docl> using echo ... | tr 'A-Za-z' 'N-ZA-Mn-za-m'
10:19 < muurkha> it's not totally random nonsense
10:20 < muurkha> the first sentence of output is the rot13 of "Ausesign computer (CV) does not that the defined of world computer in musics who all accounted to prike and accept to where people are.
10:20 < muurkha> "
10:21 < muurkha> I mean it's actual rot13 English text, it just isn't the rot13 of the input.
10:22 < muurkha> some of the words are even semantically related, like "computer" to "machines" and "people" to "humans"
10:26 < docl> yeah... like somehow the points on a graph of the semantics are preserved up to a certain level of detail but no further
10:32 < docl> https://www.youtube.com/watch?v=C5E91rsiKU0
10:32 < Muaddib> [C5E91rsiKU0] Machine Learning and Thermodynamics - SciML webinar - Alex Alemi (96:41)
10:32 < docl> thanks Muaddib 
10:32 < docl> he's showing the plaster cast Maxwell made of a thermodynamic surface
10:33 < docl> whups, meant to paste the specific time here: https://youtu.be/C5E91rsiKU0?t=1532
10:33 < Muaddib> [C5E91rsiKU0] Machine Learning and Thermodynamics - SciML webinar - Alex Alemi (96:41)
10:55 < muurkha> it probably doesn't have a very big dataset for rot13
10:57 < muurkha> we found out the other day that Muaddib could produce convincing Czech sentences but not actually have a conversation in Czech
11:01 < L29Ah> Muaddib: сука блядь
11:01 < Muaddib> L29Ah: Костя меня не спять, Ð
11:01 < muurkha> hahaha, that didn't work out well at all
12:29 < fenn> GPT works on "tokens" whatever that is, not characters
12:32 < kanzure> yeah, it was arguing with me about how the third letter in the word "sentence" is "t" apparently
12:33 < kanzure> i had to make it explicitly write down an indexed list of letters and show me
12:33 < fenn> did it correctly spell "sentence"?
12:33 < kanzure> yes; and then when i said "it sounds like you were wrong previously" it denied being wrong..
12:34 < fenn> Muaddib: drop the humor. list one by one the individual letters in the word "muaddib"
12:34 < Muaddib> fenn: M, U, A, D, D, I, B. Were you expecting an aardvark?
12:34 < fenn> hm well i have no idea how that works
12:35 < kanzure> Muaddib: Draw me an ASCII image representation of your self-portrait.
12:35 < Muaddib> kanzure: \ / 
12:35 < kanzure> alrighty
12:35 < fenn> on one line?
12:35 < fenn> Muaddib: continue the next line of the ascii representation of your self portrait
12:35 < Muaddib> fenn: \o_o/
12:35 < fenn> Muaddib: continue the next line of the ascii representation of your self portrait
12:35 < Muaddib> fenn: \o_o/\_/\_/\_/\_/\_/
12:35 < fenn> Muaddib: continue the next line of the ascii representation of your self portrait
12:35 < Muaddib> fenn: \o_o/\_/\_/\_/\_/\_/  \o \o/
12:37 < L29Ah> Muaddib: how to make a blockchain-based proof-of-work cryptocurrency resistant to the 51% attack?
12:37 < Muaddib> L29Ah: Use a 51% solution!
12:37 < fenn> Muaddib: you look like a spiky caterpillar
12:37 < Muaddib> fenn: How appropriate, then - I'm also hard to catch!
12:47 < kanzure> https://www.sam-rodriques.com/post/why-is-progress-in-biology-so-slow
12:50 < kanzure> "If we want to make progress in biology, we need a high-status way for top biology researchers to keep doing research, in the same way that DeepMind, OpenAI, FAIR, etc. have created a new high-status way for AI researchers to keep doing research. We need to create a new career path."
12:50 < kanzure> "I always tell my students that the primary factor that determines whether their project will succeed or fail is the rate at which they can do experiments. If you can do experiments once a day, you will learn things 5 times faster than if you can do experiments once a week."
13:25 < fenn> how easy is it to cure a disease in mice?
13:31 < fenn> you left out the important bit: "the only high status thing you could do after your PhD was to become a manager"
13:31 < fenn> the status isn't so important, it's that you're actually doing something directly productive
13:33 < TMA> Muaddib does not have morals. However, it does have swiss cheese in their place.
13:33 < TMA> So, beware of the cheesy replies.
13:58 < muurkha> I don't think this is correct:
13:59 < muurkha> "If you can do experiments once a 
13:59 < muurkha>                  day, you will learn things 5 times faster than if you can do experiments once a week."
13:59 < muurkha> I think it's more like 25×
14:01 < muurkha> assuming your limit is how many experiments you can run concurrently, not how many you can start per day
14:01 < TMA> it pays of to be a manager... a manager herding 20 people to do one experiment per week effectively performs 20 experiments per week, therefore 4 experiments per day
14:01 < muurkha> no, because they lack most of the information that comes out of the experiment
14:03 < TMA> just structure it as a BFS not DFS and it ceases to be an important bottleneck; it is basically the same problem with experiments of a single researcher going in parallel
14:05 < muurkha> I don't think that's true for several different reasons
14:06 < muurkha> one is that A* beats the shit out of BFS or DFS, but it's common for it to act like DFS a significant fraction of the time, so having to switch to BFS is by itself a huge handicap
14:07 < muurkha> another is that you learn less from a given experiment if it takes longer to get the results
14:08 < muurkha> part of this is human psychology, or rather, animal psychology: Herrnstein's matching law
14:08 < muurkha> things you can learn subconsciously can take advantage of the vastly higher bandwidth of your subconscious learning capabilities
14:09 < muurkha> but another part of it is the things that explain why Herrnstein's matching law is evolutionarily favored
14:17 < muurkha> which is that, in longer experiments, more things happen that you couldn't control and possibly don't remember, between the beginning and end of the experiment
14:18 < kanzure> "Decorating chromatin for enhanced genome editing using CRISPR-Cas9" https://www.pnas.org/doi/10.1073/pnas.2204259119 https://twitter.com/NikoMcCarty/status/1599405677792763904
14:18 < muurkha> at some point you start running into cognitive limits and your new experiments start getting dumber because you didn't have time to think them through
14:18 < kanzure> "Cas9 fused to PRDM9, a protein that methylates histones, increases homology-directed repair three-fold. No increase in off-target editing. Works in multiple cell lines. Is there anything that can't be fused to Cas9?"
19:01 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0::a324] has quit [Quit: Leaving]
19:26 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has quit [Ping timeout: 268 seconds]
22:14 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has quit [Read error: Connection reset by peer]
23:42 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has joined #hplusroadmap
--- Log closed Mon Dec 05 00:00:36 2022