--- Log opened Sat Jul 01 00:00:46 2023
01:02 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has joined #hplusroadmap
01:41 -!- TMM_ [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
01:42 -!- TMM_ [hp@amanda.tmm.cx] has joined #hplusroadmap
03:27 < alethkit> My point is similar to Taleb's
03:28 < alethkit> That it's less about the specific confidence interval, and more about the payoff associated with any errors in your model
04:51 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has left #hplusroadmap []
05:00 < hprmbridge> kanzure> synthetic training data https://arxiv.org/abs/2305.07759
06:19 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0:550:e1ce:52d0:412c] has joined #hplusroadmap
06:50 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
06:59 -!- A_Dragon [A_D@libera/staff/dragon] has quit [Quit: ZNC - https://znc.in]
07:00 -!- A_Dragon [A_D@libera/staff/dragon] has joined #hplusroadmap
07:09 -!- A_Dragon [A_D@libera/staff/dragon] has quit [Read error: Connection reset by peer]
07:13 -!- A_Dragon [A_D@libera/staff/dragon] has joined #hplusroadmap
07:20 -!- test_ [flooded@gateway/vpn/protonvpn/flood/x-43489060] has joined #hplusroadmap
07:24 -!- flooded [flooded@gateway/vpn/protonvpn/flood/x-43489060] has quit [Ping timeout: 250 seconds]
08:28 < muurkha> alethkit: I feel like a Bayesian agent can gracefully handle errors in any specific model by calculating a nonzero probability that that model is wrong
08:31 < muurkha> 16% the die will come up 1, 16% it will come up 2, ... 16% it will come up 6, 3% something quotidian will interfere with rolling it (like it falls off the table and gets lost), 0.9% chance of something extraordinary interfering with rolling it (a loaded die, a plane crash), 0.1% chance my understanding of the situation is totally incorrect and the die doesn't even exist and I'm dreaming or something
08:59 < jrayhawk> which is a good out for fermi-estimation-based pascal's muggings
09:05 < alethkit> muurkha: But where are you pulling those numbers from?
09:16 < muurkha> alethkit: you seem to be asking where priors come from.  there isn't a good answer to that; Bayes' Theorem only describes how to update probabilities, not how to derive them from nothing
09:16 < muurkha> pascal's mugging is always a problem
09:19 < muurkha> alethkit: I might suggest reading an introduction to Bayesian epistemology such as https://plato.stanford.edu/entries/epistemology-bayesian/#BayeDiviProbPrio
09:20 < kanzure> .wik retatrutide
09:21 < EmmyNoether> HTTPError: HTTP Error 404: Not Found (wik:100)
11:54 < mrdata> if you dont have priors, begin with bayesian indifference
12:53 < muurkha> it'd be great if that was a sensible answer.  up for a game of reference class tennis?
12:59 < fenn> i was reading the tinystories synthetic training data paper last night. this is the sequel, "textbooks are all you need" focused on coding https://arxiv.org/pdf/2306.11644.pdf
13:01 < fenn> it turned out not to be what i expected
13:02 < fenn> the tinystories idea is that you don't want to overwhelm the language model with lots of irrelevant facts, let it concentrate on building up a model of grammar and the basics of how the world works. naturally the sequel to that should be, okay once we have grammar and a basic model of the world, we can add more neurons and more data about the world, gradually increasing its training data in a way
13:02 < fenn> that simple concepts build to advanced concepts, the way we teach humans
13:03 < fenn> but instead they're just like "it's all about data quality!!!"
13:03 < fenn> maybe an unfair assessment, i haven't really dug into the paper yet
13:11 < fenn> phi-1 gets 51% on HumanEval, a code writing/understanding/debugging benchmark. if there aren't serious systematic errors, that should be good enough to get good results with a tree of thoughts style prompting
13:12 < fenn> not bad for one GPU-month of training
13:39 -!- flooded [flooded@gateway/vpn/protonvpn/flood/x-43489060] has joined #hplusroadmap
13:42 -!- test_ [flooded@gateway/vpn/protonvpn/flood/x-43489060] has quit [Ping timeout: 240 seconds]
14:14 -!- TMM_ [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
14:14 -!- TMM_ [hp@amanda.tmm.cx] has joined #hplusroadmap
16:11 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has quit [Ping timeout: 240 seconds]
16:33 < L29Ah> https://www.tomshardware.com/news/chinese-researchers-usedai-to-design-industrial-scale-risc-v-cpu-in-under-5-hours
16:48 -!- test_ [flooded@gateway/vpn/protonvpn/flood/x-43489060] has joined #hplusroadmap
16:51 -!- flooded [flooded@gateway/vpn/protonvpn/flood/x-43489060] has quit [Ping timeout: 250 seconds]
17:00 -!- Netsplit *.net <-> *.split quits: Croran, archels, gwillen, Betawolf
17:00 -!- archels [~neuralnet@159.69.156.65] has joined #hplusroadmap
17:00 -!- Netsplit over, joins: Betawolf
17:00 -!- gwillen [gwillen@2600:3c01::f03c:92ff:fe31:d7fb] has joined #hplusroadmap
17:02 -!- gwillen [gwillen@2600:3c01::f03c:92ff:fe31:d7fb] has quit [Changing host]
17:02 -!- gwillen [gwillen@user/gwillen] has joined #hplusroadmap
17:15 -!- Croran [~Croran@user/Croran] has joined #hplusroadmap
19:03 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0:550:e1ce:52d0:412c] has quit [Quit: Leaving]
19:18 -!- justanotheruser [~justanoth@gateway/tor-sasl/justanotheruser] has joined #hplusroadmap
19:28 -!- TMM_ [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
19:29 -!- TMM_ [hp@amanda.tmm.cx] has joined #hplusroadmap
20:47 -!- uzkruh [~uzkruh@fl-67-235-216-185.dhcp.embarqhsd.net] has joined #hplusroadmap
20:49 < uzkruh> Will accumulation of capital, pharmaceuticals, biotech, and AI eventually justify civilization?
21:25 -!- deltab [~deltab@user/deltab] has quit [Ping timeout: 245 seconds]
21:30 -!- uzkruh [~uzkruh@fl-67-235-216-185.dhcp.embarqhsd.net] has quit [Quit: Client closed]
21:30 < hprmbridge> nmz787> Accumulation by who? And is justification of civilization even a valid question?
21:31 < hprmbridge> nmz787> It seems like a product of evolution to me. Does evolution get justified???
21:35 -!- deltab [~deltab@user/deltab] has joined #hplusroadmap
21:41 -!- mxz [~mxz@user/mxz] has quit [Quit: cya]
21:42 -!- mxz [~mxz@user/mxz] has joined #hplusroadmap
21:43 < fenn> we are survival machines
21:44 < fenn> if civilization kills everyone, it turns out not to have been justified
21:44 < fenn> the question is not even wrong
21:51 < fenn> this is what your god-king answered...
21:51 < fenn> Techno-optimist: Yes, advancements in capital, pharmaceuticals, biotech, and AI are the driving forces of civilization. They improve living standards, extend life expectancy, and open up new possibilities for human potential.
21:51 < fenn> Social justice advocate: No, these advancements alone can't justify civilization. They often exacerbate inequality and marginalize certain groups. A just civilization must prioritize fairness and inclusivity.
21:52 < fenn> Environmentalist: No, these advancements often come at a great cost to our planet. A sustainable civilization must prioritize ecological balance and respect for all life forms.
21:52 < fenn> Humanist: These advancements are tools, not ends in themselves. A civilization is justified by its ability to promote human flourishing in all its dimensions, including moral, intellectual, and emotional growth.
21:52 < fenn> Existentialist: Civilization's justification lies in its ability to provide individuals with the freedom to create their own meaning and purpose, which can't be reduced to technological or economic progress.
21:56 < fenn> on the other hand, storing our genetic sequences as redundant error correction coded digital data is an improvement over soupy muck constantly at risk of being torn to shreds by oceans of hostile R-strategists
21:57 < fenn> oh i am talking to nobody again
23:06 -!- flooded [flooded@gateway/vpn/protonvpn/flood/x-43489060] has joined #hplusroadmap
23:10 -!- test_ [flooded@gateway/vpn/protonvpn/flood/x-43489060] has quit [Ping timeout: 264 seconds]
--- Log closed Sun Jul 02 00:00:47 2023