--- Log opened Fri Dec 16 00:00:46 2022
00:55 < fenn> https://www.riffusion.com/about  stable diffusion trained on audio spectrograms works really well for music generation
00:57 < fenn> griffin-lim sounds really harsh but this can be fixed with a small convnet
06:21 < kanzure> it's common to see people commenting on social media about how they have xyz  horrible genetic disease but they are glad they are alive
06:22 < kanzure> i wonder if the proportion of genetic horror commentators is higher than the proportion of IVF baby grateful commentators thankful for being alive due to IVF
06:22 < kanzure> or if there is a lopsided representation/selection effect
06:22 < kanzure> like, i doubt that IVF babies are unhappy to be alive, but they might be less likely to comment about the whole thing
06:23 < kanzure> birth defects are 1/33 and IVF is only 1/100 births
07:07 < kanzure> anyway, a retrospective on media/academic response would be useful
09:01 -!- TMM_ [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
09:01 -!- TMM_ [hp@amanda.tmm.cx] has joined #hplusroadmap
09:47 -!- juri_ [~juri@84-19-175-179.pool.ovpn.com] has quit [Ping timeout: 268 seconds]
10:09 < nmz787> fenn: riffusion is neat, but my attempts to use it have produced terrible results
10:09 < nmz787> I wonder if they specifically didn't train it with artist names and song titles
10:10 < nmz787> even something more generic 'bollywood banjos' sounds more like latin salsa
10:11 < nmz787> 'bangra banjos' sounds like jazz....
10:11  * nmz787 cries about how the AI model is biased!
10:12 < nmz787> s/cries/screams/
10:12 -!- juri_ [~juri@84-19-175-179.pool.ovpn.com] has joined #hplusroadmap
10:13 < kanzure> for music generation my choice would be to train on MIDI files
10:14 < kanzure> and create an abstract syntax tree for setting up the instruments and melodies or something; it's not like there's huge variations in structure.
10:14 < kanzure> (music should be considered more like programming/openai codex)
10:15 < nmz787> none of the outputs I'm getting make any sense
10:15 < nmz787> 'tom waits white christmas' gives some jazz sounding crap
10:16 < kanzure> and any generator should just use the conventional soundfonts that are already available
10:16 -!- Festive_Dragon [A_D@libera/staff/dragon] has quit [Ping timeout: 633 seconds]
10:17 < nmz787> 'gravely voice singing happy birthday' again more jazz... sounds nothing like happy birthday song, voice is high pitched and "clear" (though unintelligible)
10:19 < nmz787> maybe I just suck at talking to AI
10:37 -!- cthlolo [~lorogue@77.33.23.154.dhcp.fibianet.dk] has joined #hplusroadmap
10:38 < docl> really hard to get chatgpt to provide correct citations with doi. It prefers to make up fake ones or pick them at random
10:46 -!- A_Dragon [A_D@libera/staff/dragon] has joined #hplusroadmap
10:56 < muurkha> well of course
11:00 < kanzure> A_Dragon: welcome.
11:55 -!- A_Dragon is now known as Festive_Dragon
12:42 < fenn> it probably wasn't trained on any coprighted music, dunno
12:42 < fenn> it's a miracle that riffusion works at all
12:43 < fenn> they're literally using the stable diffusion image generation model
12:44 < fenn> personally i don't know what "bangra" refers to, so i can't blame the AI for not knowing either
12:44 < fenn> i mean, i can guess
12:45 < fenn> i can't run the riffusion web app because of webgl
12:45 < fenn> and other reasons
12:47 < fenn> can you upload an audio sample for it to riff on?
13:00 < kanzure> chatGPT is absolutely incapable of talking about he jiankui's work without believing all the talking points from the media, it's really fascinating to watch
13:02 < kanzure> https://diyhpl.us/~bryan/irc/chatgpt-he-jiankui.txt
13:11 < lsneff> .tw 1603826876052656129
13:11 < saxo> yes, excited spacex employee, this includes you too (@whitequark, in reply to tw:1603826504072298515) #BoycottTwitter
13:12 < lsneff> very good change whitequark is talking about me. i respect her a lot, so this hurts to hear
13:12 < lsneff> *chance
13:18 < kanzure> i wouldn't pay any attention to this until later; it's natural for the dying media/journalism class to have their last kicks before their death
13:19 < kanzure> it would actually be more surprising if they weren't complaining at all
13:42 < muurkha> "bangra" is presumably a misspelling of "bhangra"
13:42 < muurkha> lsneff: condolences
13:52 < L29Ah> https://storage.gra.cloud.ovh.net/v1/AUTH_011f6e315d3744d498d93f6fa0d9b5ee/qotoorg/cache/media_attachments/files/109/525/379/297/073/847/original/3657a705f7d0250a.png
13:53 < muurkha> whitequark can be aggressive sometimes, and these are especially polarized times, thanks to things like the Ukrainian war and cryptocurrencies
14:41 < kanzure> https://agingbiotech.info/companies/static.html
15:09 < kanzure> "As far as we know, He has not published noteworthy scientific papers in the gene-editing field and was not actively involved in the gene editing community in China. We were enraged by this extremely irresponsible misconduct."
15:10 < kanzure> from https://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.3000224
15:14 < kanzure> 'global scientific consensus' aka dogma
15:22 < kanzure> chinese academy of sciences seems to have removed their 2018 condemnation?
15:22 -!- Jenda [~jenda@coralmyn.hrach.eu] has quit [Remote host closed the connection]
15:26 < kanzure> "与此同时这对于中国生物医学研究领域在全球的声誉和发展都是巨大的打击，对中国绝大多数勤勤恳恳科研创新又坚守科学家道德底线的学者们是极为不公平的。
15:26 -!- Jenda [~jenda@coralmyn.hrach.eu] has joined #hplusroadmap
15:27 < kanzure> "joint statement of scientists" https://tech.qq.com/a/20181126/014341.htm can't find the original weibo source
15:28 -!- ANACHRON [~Malvolio@idlerpg/player/Malvolio] has quit [Ping timeout: 252 seconds]
16:04 -!- cthlolo [~lorogue@77.33.23.154.dhcp.fibianet.dk] has quit [Read error: Connection reset by peer]
16:15 < kanzure> my my what an over-developed clown court: "Scientific works are normally published in peer-reviewed journals, but He failed to do so regarding the birth of gene-edited babies. This was one of the grounds on which He was criticized.[29][104] It was later reported that He did submit two manuscripts to Nature and the Journal of the American Medical Association, which were both rejected, mainly on ...
16:15 < kanzure> ...ethical issues.[105] He's first manuscript titled "Birth of Twins After Genome Editing for HIV Resistance" was submitted to Nature on 19 November."
16:16 < kanzure> crazy they would reject a paper about an experiment that has already occurred for ethical issues, and then criticise him for not having published. kafkaesquean horror show.
16:19 < muurkha> perhaps the people who were criticizing him for not having published didn't know that his manuscripts had been rejected
16:45 < kanzure> doubtful, he had a publication that was accepted and published and then they complained until it was taken down
16:46 < muurkha> are you sure it was the same people?
16:47 < kanzure> "scientists" yes
16:48 < kanzure> the existence of clinical work should not impact whether you are allowed to think about and publish about ethics https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6383508/
16:52 < muurkha> concur
16:52 < muurkha> maybe they were different "scientists".  I'm always suspicious of anyone who attributes an action or belief to "scientists"
16:52 < kanzure> ( https://www.liebertpub.com/doi/10.1089/crispr.2018.0051 )
19:18 -!- ANACHRON [~Malvolio@idlerpg/player/Malvolio] has joined #hplusroadmap
20:35 < L29Ah> https://www.lifespan.io/news/short-bouts-of-vigorous-activity-may-reduce-mortality-risk/ is it me or the cause and effect there are swapped?
20:47 < fenn> a little of A, a little of B
21:08 -!- ANACHRON [~Malvolio@idlerpg/player/Malvolio] has quit [Ping timeout: 268 seconds]
21:12 < fenn> "If the web were invented a few years ago, today they'd be putting Tim Berners-Lee in Fortune and asking him why his new app didn't tell you which websites to look at and how he planned to get rid of misinformation"
21:18 < superkuh> The web was re-invented a few years ago as HTTP/3 (QUIC over UDP). The spec can be interpreted many ways, but almost all implementations do not allow connections to any address without that site having a CA TLS cert. So there's the mechanism of control.
21:20 < fenn> yes i ran into this yesterday
21:20 < fenn> couldn't connect to a friend's server because it was self-signed and used HSTS
21:29 < muurkha> L29Ah: that was my thought, too
21:30 < muurkha> superkuh: I didn't realize that!
21:34 < superkuh> I forget who said this on #web, but: "So wow, this may well be the single sleaziest thing I have ever seen a hosting provider do: An HSTS ransom. The host charges (a LOT) extra to allow HTTPS. In 2022 lol. They gave the client an unannounced free trial of that feature. And set this header: Strict-Transport-Security: max-age=63072000; includeSubDomains; preload"
21:57 -!- ANACHRON [~Malvolio@idlerpg/player/Malvolio] has joined #hplusroadmap
22:00 < muurkha> wow, that's lawsuit material
22:15 < L29Ah> just swap the hosting provider duh
22:18 < muurkha> the new hosting provider won't have access to the necessary certificate to satisfy HSTS
22:19 < L29Ah> HSTS doesn't pin the certificate IIRC
22:19 < muurkha> isn't that precisely what HSTS does?
22:20 < L29Ah> https://en.wikipedia.org/wiki/HTTP_Strict_Transport_Security 0 mentions of certificate pinning
22:23 < muurkha> I wonder if I'm confusing HSTS with HPKP
22:24 < fenn> i think it just sets a bit that says you must use HTTPS from now on, so no downgrade to HTTP attacks
22:25 < muurkha> that seems to be correct
23:11 < sknebel> yep. key pinning was HPKP and isnt supported by browsers anymore since ~2020
23:11 < sknebel> due to the misuse and footgun risk
23:15 -!- ANACHRON [~Malvolio@idlerpg/player/Malvolio] has quit [Read error: Connection reset by peer]
--- Log closed Sat Dec 17 00:00:47 2022