--- Log opened Mon Mar 13 00:00:09 2023
01:33 -!- zhx [~zhx@103.167.99.243] has joined #hplusroadmap
01:33 -!- zhx [~zhx@103.167.99.243] has quit [Client Quit]
03:19 -!- luna__ [~luna@82-132-228-20.dab.02.net] has joined #hplusroadmap
03:21 -!- luna_ [~luna@user/luna/x-4729771] has quit [Ping timeout: 252 seconds]
03:23 -!- luna__ [~luna@82-132-228-20.dab.02.net] has quit [Read error: Connection reset by peer]
03:29 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has quit [Read error: Connection reset by peer]
03:56 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
05:37 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0:854e:cd7c:7359:37f2] has joined #hplusroadmap
05:39 < hprmbridge> kanzure> http://web.archive.org/web/20130513043046/http:/finney.org/~hal/pol_v_tech.html
05:42 < hprmbridge> SarahC> Inducing meiosis?
05:42 < hprmbridge> SarahC> https://denovo.substack.com/p/meiosis-is-all-you-need
05:43 < hprmbridge> kanzure> yeah, iterated embryo selection only makes sense if you don't know about genetic engineering
05:43 < hprmbridge> SarahC> I think Guzey’s org is not that well funded so I’d stick to projects that  cost <$1M, which suggests sticking to in vitro
05:45 < hprmbridge> kanzure> even this proposal is a little odd; why not just.... use genetic engineering.
05:55 < hprmbridge> yashgaroth> newscience are indeed not well-funded. IES (or this version of it) has benefits over genetic engineering until you can link SNPs to causative mutations. But that's a lot of genome sequencing, and keeping cells in vitro for that long, especially haploid, means you'll probably get too many chromosomal aberrations before you get a final cell that has all the SNPs you want
05:58 < hprmbridge> yashgaroth> I suppose for pure IES you'd only need SNP chips instead of full sequencing which would save some costs
06:26 < hprmbridge> yashgaroth> also the expense of splitting tens of thousands of meiotic colonies to find one that has all of a parent's best SNPs seems like a lot of effort for limited gain. It won't mollify the bioethicists just because no "editing" is happening (assuming Merrick finds a way to do the haploidization with transient expression only). Versus, say, editing an embryo-derived ESC line with whatever hundreds of
06:26 < hprmbridge> yashgaroth> mutations we want, sequencing monoclonal colonies, and nuclear-transferring a fully genome sequenced cell into an oocyte
06:29 < hprmbridge> yashgaroth> the benefit of IES is that you're doing hundreds of iterated cycles with hundreds of pre-selected parents to generate the perfect ur-child with all possible beneficial SNPs contained in one genome, who immediately has a grand mal after being born
06:29 < hprmbridge> msnewgooty> It’s impractical…but so is introducing a bunch of SNPs (currently). You want to do 100-plex prime editing?
06:30 < hprmbridge> kanzure> there's too much focus on SNPs. you can insert entire genes or circuits that do what you want.
06:31 < hprmbridge> kanzure> i think the record for cas9 edits is in the many thousands? from church lab multiplexing machine.
06:32 < hprmbridge> yashgaroth> pretty much. Crack open an embryo, grow the cells in a dish, blast them with all kinds of editing, then plate out single cells that grow into colonies. You can sequence colonies to find one that has all the mutations (or inserts or neochromosomes or whatever), then pick a cell from the colony and pop the nucleus into an egg
06:32 < hprmbridge> msnewgooty> The cas9 multiplexing record isn’t really that
06:33 < hprmbridge> kanzure> yeah I do t have a reference in front of me
06:33 < hprmbridge> kanzure> I'll have to double check
06:33 < hprmbridge> msnewgooty> Luhan targeted a repetitive perv
06:33 < hprmbridge> msnewgooty> so of course it will cut a bajillion times
06:33 < hprmbridge> msnewgooty> But it’s a single guide
06:33 < hprmbridge> msnewgooty> Also if you ever made that many DSBs in a human CBER would flip
06:33 < hprmbridge> msnewgooty> For good reason
06:34 < hprmbridge> yashgaroth> you can do it in cycles, find a colony that has the mere dozen edits you did and then edit cells from that colony and repeat the procedure
06:34 < hprmbridge> kanzure> anyway, lots of interesting things are a single mutation or you just inset a whole new gene/circuit
06:34 < hprmbridge> msnewgooty> There’s a reason all proposed multiplexed knockout pipelines are with CBE, less chance of translocation
06:34 < hprmbridge> msnewgooty> The circuit approach is compelling
06:35 < hprmbridge> kanzure> https://diyhpl.us/wiki/genetic-modifications/
06:35 < hprmbridge> yashgaroth> anyway, all the IES in the world won't get you a DEC2 short-sleep mutant and that alone is enough for people to consider editing (CBER aside)
06:37 < hprmbridge> kanzure> a few days ago ansabio reported the first 1000mer synthesized by TdT chemistry
06:37 < hprmbridge> SarahC> What about just removing all rare variants? That doesn’t require understanding how anything works and will probably lead to improvements on massively polygenic traits
06:37 < hprmbridge> kanzure> I like to call that a "consensus genome"
06:37 < hprmbridge> SarahC> Yep
06:37 < hprmbridge> kanzure> bram cohen (creator of BitTorrent) is a big fan of that approach
06:37 < hprmbridge> yashgaroth> what's wrong with rare variants, who wants the most average possible child
06:38 < hprmbridge> kanzure> One reason might be to minimize the chance of defect. But if you really wanted to do that, you should just consider a cloning yourself.
06:38 < hprmbridge> yashgaroth> (and yes with true IES I guess you could include some short-sleep mutant carriers in the parent pool, I'm just talking about two parents having a kid)
06:39 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has joined #hplusroadmap
06:40 < hprmbridge> SarahC> (Theory goes that most rare variants slightly degrade things like brain performance and most people have *some* — a consensus genome wouldn’t be an average person but an extraordinary one, same way if you average many faces you get an unusually beautiful face.)
06:43 < hprmbridge> kanzure> alright well, let's do it
06:49 < hprmbridge> yashgaroth> idk how much of the face-averaging attractiveness is just due to removal of asymmetry and other minor flaws, or unusually large/small features. Sadly not really testable unless we convince biobanks to start rating peoples' looks
06:54 < hprmbridge> yashgaroth> anyway I don't think we'd need hundreds of edits to reach the limit of high-confidence beneficial mutations in a cell, especially beyond what could be done (a little more) safely with base editing. Almost all GWAS SNPs are still just correlative, there's no point editing a SNP unless it's a functional protein mutation
07:50 < hprmbridge> msnewgooty> It would be interesting to see these as a table with tissue (especially liver or ex vivo) and type of mutation needed to be introduced (KD, base edit, insertion, etc)
07:50 < hprmbridge> msnewgooty> aside from CCR5, I wonder how many could be done ex vivo
07:51 < hprmbridge> kanzure> One of the other things that we will need to do eventually for this document is make it very clear what the p-values are... not everything there is as "real" as other changes.
07:51 < hprmbridge> kanzure> some are more real than others
07:53 < hprmbridge> kanzure> I also want to make a "mutantpedia" encyclopedia of human mutations and especially familial mutations (like the short sleepers and pain insensitivity families)
07:53 < hprmbridge> kanzure> mutantpedia would also directly reference the people and their genetic material, so that we could go get it or do more analysis
07:56 < hprmbridge> msnewgooty> Like a wiki version of OMIM
07:56 < hprmbridge> kanzure> hm?
07:56 < hprmbridge> msnewgooty> https://www.omim.org/about
07:56 < hprmbridge> msnewgooty> That's the initial go-to for looking at variants in a disease, in my experience
07:59 < hprmbridge> kanzure> btw what is your background?
08:01 < hprmbridge> kanzure> in an ~hour doing a call with these guys https://finalspark.com/artificial-intelligence-project-status/ let me know if anyone has questions they would like me to ask
08:03 < hprmbridge> msnewgooty> molecular biology, I guess
08:04 < hprmbridge> kanzure> that's cool
08:06 < hprmbridge> msnewgooty> re: insertion of genes, have you ever talked to the Rejuvenate Bio crowd? Ex-Church lab
08:07 < hprmbridge> msnewgooty> not genomic insertion, per se. But episomal expression via AAV is along those lines
08:08 < hprmbridge> kanzure> not me personally, but I'd be open to it.
08:08 < hprmbridge> kanzure> I don't have much interest in viral gene therapy except for research and maybe surface level somatic therapies in human. I just don't see how to get delivery to all cells in the body... plus most of the fun stuff has to happen during fetal development...
08:10 < hprmbridge> msnewgooty> Fair, though probably the quickest way to test the effects of some of these things is a somatic therapy
08:10 < hprmbridge> kanzure> msnewgooty: you might be interested in our enzymatic DNA synthesis subgroup (basically the ppl in this discord but we wrote it down in emails) https://groups.google.com/g/enzymaticsynthesis/c/41r7D3wHKn8
08:11 < hprmbridge> kanzure> or this one https://groups.google.com/g/enzymaticsynthesis/c/r1lUxyoDWE4
08:12 < hprmbridge> msnewgooty> Oh cool. I learned about https://averybio.com/ recently. It's more of the parallel route (Twist competitor?) but making oligo libraries at 1M+ scale for <$100 would be compelling
08:13 < hprmbridge> kanzure> we have an unpublished DNA synthesis technique that we've been working on
08:14 < hprmbridge> kanzure> we want 10,000+ bp in single molecules no ligation or assembly
08:15 < hprmbridge> msnewgooty> that would be compelling, if speed/scalability/price were solved accordingly
08:17 < hprmbridge> msnewgooty> you can order 1500-mers (arrayed, not pooled; dsDNA) from IDT at 5c/base right now with ~4 day turnaround via eBlocks. But that's pricier than preferred and also they have restrictions on sequence (because they are cloning or some trick)
08:20 < hprmbridge> kanzure> do you know about ribbon biolabs?
08:20 < hprmbridge> kanzure> they do 3192-plate of oligos, probably a stack of them, and then they do pick and place assembly ligation of different molecules from the oligos
08:22 < hprmbridge> kanzure> the plate seems to be synthesized by conventional oligonucleotide synthesis chemistry. presumably there is some speedup because they can computationally choose the ideal fragments and order them so that the ones most used are on the 1st plate... etc.
08:28 < hprmbridge> msnewgooty> if they're doing pick and place from a 5-mer plate, are they going to have a robotic station for each bot?
08:29 < hprmbridge> msnewgooty> with phosphoramidite synthesizers, you can stuff rows upon rows of those in a warehouse
08:29 < hprmbridge> kanzure> their materials suggest more like 10 to 30mers
08:29 < hprmbridge> msnewgooty> the footprint/TAT/Opex per station would be rough
08:29 < hprmbridge> kanzure> They might be using a robotic fluid arm or they might be doing microfluidics. I don't really know.
08:30 < hprmbridge> msnewgooty> I mean, in concept, it's compelling -- if you could do the phosphoramidite 10 at a time rather than 1 at a time, you instantly get that win (not sure about purification tho). But the scaleup isn't clear to me immediately
08:31 < hprmbridge> kanzure> At the high end you can imagine a stack of 3192 plates and each spot has a unique oligo. For a large construct this would be too many plates. So I think they just reuse oligos and split up the construct so that it consists of multiple repeating subfragments....
08:31 < hprmbridge> msnewgooty> there's a big difference about doing something bespoke versus productized that I think need to be baked in. Also how to make money off of it (not my problem, but could be theres)
08:31 < hprmbridge> msnewgooty> Like many investors passed on Ansa because the economics don't make sense. They're going to become a therapeutics company. Because they have to
08:31 < hprmbridge> kanzure> you saw the ansa announcement the other day?
08:32 < hprmbridge> msnewgooty> yeah
08:32 < hprmbridge> msnewgooty> most syn bio people I know met it with a yawn
08:32 < hprmbridge> kanzure> I've been wondering if they just use a long time per reaction to get that high of a yield per step
08:32 < hprmbridge> msnewgooty> I've never played with the TdT things myself
08:33 < hprmbridge> kanzure> are you in the industry? academia?
08:42 < hprmbridge> msnewgooty> Primarily academic. Co founded companies but not full time
09:30 -!- Llamamoe [~Llamamoe@46.204.68.7.nat.umts.dynamic.t-mobile.pl] has quit [Quit: Leaving.]
09:43 < hprmbridge> nmz787> Which?
09:43 < hprmbridge> nmz787> I really want to get back onto dna lab stuff
09:45 < hprmbridge> nmz787> Now my wife wants a (not necessarily ) new vehicle, and I'm not sure I feel good spending money on my lab improvements right now.
09:46 < hprmbridge> nmz787> Well, that coupled with wanting a tractor and big ass wood chipper, and paying a company to come mow and/or spray herbicide around the timber seedling area
09:47 < hprmbridge> nmz787> Although I'm somewhat inclined to let that last one slide, in lieu of battling it out with nature myself
09:48 < hprmbridge> nmz787> I think we did the last task at our to-be rental property yesterday, so I think that will be generating income soon and be mostly over with in terms of time investment
09:48 < hprmbridge> nmz787> Feels like my weekends have been piddling around at that place for almost two years
10:06 < kanzure> antonio regalado v cody sheehy on twitter in a bit https://twitter.com/MakePeople_Film/status/1635320007046610944
10:08 < hprmbridge> kanzure> https://twitter.com/johnedgarpark/status/1635086070936395776
10:08 < kanzure> ("i found this note inside of a synthesizer i'm fixing up")
10:09 < kanzure> wrong synthesizer type, nevermind.
10:23 < kanzure> https://www.amazon.com/Natural-General-Intelligence-understanding-brain/dp/0192843885 from the summerfield lab https://humaninformationprocessing.com/publications/
10:24 < kanzure> i don't really have a good handle on this group https://wba-initiative.org/en/ but a few different links keep leading to their whole brain architecture working group
10:34 -!- flooded is now known as _flood
10:48 -!- Gooberpatrol66 [~Gooberpat@user/gooberpatrol66] has joined #hplusroadmap
10:57 < kanzure> "This is allegedly a Morgan Stanley note on GPT-4/5 training demands, inference savings, Nvidia revenue, and LLM economics:  ... “We think that GPT 5 is currently being trained on 25k GPUs - $225 mm or so of NVIDIA hardware…”" https://twitter.com/XiXiDu/status/1635250014162391042 https://www.reddit.com/r/mlscaling/comments/11pnhpf/morgan_stanley_note_on_gpt45_training_demands/ 
11:05 -!- cthlolo [~lorogue@77.33.23.154.dhcp.fibianet.dk] has joined #hplusroadmap
11:12 -!- Gooberpatrol66 [~Gooberpat@user/gooberpatrol66] has quit [Ping timeout: 246 seconds]
11:17 < kanzure> a blog series that is sort of anti-eliezer https://aiascendant.substack.com/p/extropias-children-chapter-1-the-wunderkind (honestly it's nothing new and not that interesting- it has 8 parts for some reason. lots of words.)
11:18 < kanzure> uh, 7 parts.
11:25 -!- Gooberpatrol66 [~Gooberpat@user/gooberpatrol66] has joined #hplusroadmap
11:29 -!- Gooberpatrol66 [~Gooberpat@user/gooberpatrol66] has quit [Ping timeout: 250 seconds]
11:59 < kanzure> the "make people better" documentarians are hosting a twitter spaces ~now-ish https://mobile.twitter.com/MakePeople_Film/status/1634267299220955136
12:07 < muurkha> nmz787: having a vehicle may be helpful in moving equipment into your lab
12:07 < muurkha> maybe getting it cheaper
12:09 < hprmbridge> nmz787> I have multiple vehicles 🙂 varying levels of lab equipment can be moved
12:09 < hprmbridge> nmz787> the equipment is all here, now
12:10 < kanzure> hm they are late to the twitter spaces. maybe someone else got disappeared.
12:10 < kanzure> supposedly discussing https://diyhpl.us/wiki/transcripts/makepeoplebetter/episode-001/
12:10 < hprmbridge> SarahC> Btw does anyone have a directory of active projects that need funding? (Or things ppl on this Discord are working on)?
12:10 < hprmbridge> nmz787> I just need to insulate the walls/ceiling of the lab space, install mini split heat/cooling, install bathroom and hook up the drain to a macerating waste pump (to ship it to my septic tank), build the actual cleanroom
12:11 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has quit [Read error: Connection reset by peer]
12:12 < hprmbridge> nmz787> having someone come as a "visiting scientist" at least requires the workshop bathroom to be made, and a small living space cordoned off... or a small cabin be built (possibly relying on the workshop water/waste system/room)
12:15 < kanzure> SarahC: no active directory but can ask around to see what people are up to- there's DNA synthesis, embryo editing, molecular nanotechnology (atomically precise manufacturing), some cryonics people, uh.. a few others.
12:16 < kanzure> ok here it is https://twitter.com/samira_kiani1/status/1635358409523605506
12:27 -!- Gooberpatrol66 [~Gooberpat@user/gooberpatrol66] has joined #hplusroadmap
12:29 -!- Gooberpatrol66 [~Gooberpat@user/gooberpatrol66] has quit [Remote host closed the connection]
12:29 -!- Gooberpatrol66 [~Gooberpat@user/gooberpatrol66] has joined #hplusroadmap
12:37 < lsneff> @kanzure is there going to be a transcript/recording?
12:55 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
13:03 < hprmbridge> nmz787> https://twitter.com/maxhodak_/status/1635330756276015105?s=20
13:03 < hprmbridge> nmz787>  https://cdn.discordapp.com/attachments/1064664282450628710/1084929701933498428/Screenshot_20230313-125852.png
13:03 < hprmbridge> nmz787>  https://cdn.discordapp.com/attachments/1064664282450628710/1084929711890768072/Screenshot_20230313-125921.png
13:03 < hprmbridge> nmz787> No pricing, maybe they'll approve my account then I can see
13:04 -!- cthlolo [~lorogue@77.33.23.154.dhcp.fibianet.dk] has quit [Read error: Connection reset by peer]
13:06 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has quit [Read error: Connection reset by peer]
13:15 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=623b8f45 Bryan Bishop: make people better twitter spaces >> http://diyhpl.us/diyhpluswiki/transcripts/makepeoplebetter/2023-03-13-makepeoplebetter-twitter-spaces/
13:18 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=880c52b7 Bryan Bishop: tweeter link >> http://diyhpl.us/diyhpluswiki/transcripts/makepeoplebetter/2023-03-13-makepeoplebetter-twitter-spaces/
13:19 < kanzure> lsneff: there you go^
13:24 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=97bb5ff6 Bryan Bishop: flesh out my comments >> http://diyhpl.us/diyhpluswiki/transcripts/makepeoplebetter/2023-03-13-makepeoplebetter-twitter-spaces/
13:26 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=656ae557 Bryan Bishop: clarify my comment on his reappearance >> http://diyhpl.us/diyhpluswiki/transcripts/makepeoplebetter/2023-03-13-makepeoplebetter-twitter-spaces/
13:31 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=0b7555f3 Bryan Bishop: can't talk and type at the same time :( >> http://diyhpl.us/diyhpluswiki/transcripts/makepeoplebetter/2023-03-13-makepeoplebetter-twitter-spaces/
13:32 < kanzure> sorry for all the edits; maybe it's time to put the notifications on a once-in-a-while timer instead of a firehose
13:33 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
13:35 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has quit [Read error: Connection reset by peer]
13:41 < hprmbridge> nmz787> At least they don't make it to discord
13:50 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
13:50 -!- codaraxis [~codaraxis@user/codaraxis] has joined #hplusroadmap
13:54 < lsneff> thanks
14:25 < fenn> https://info.addgene.org/hubfs/CRISPR_101_ebook/3rd%20edition/CRISPR-eBook-3rd-edition.pdf
14:31 < fenn> wouldnt it make more sense to deliver anti-crispr protein mRNA into the nucleus along with the crispr itself, so the anti-proteins get synthesized immediately after the crispr gets to work, limiting off-target activity without requiring two vectors
14:36 < fenn> or some sort of genetic regulatory circuit that acts as a timer
14:38 < yashgaroth> you still want to give the crispr time to cut, otherwise simultaneous co-delivery of an inhibitor just makes it work worse generally
14:41 < fenn> this book says 50% of the on-target activity happens in 6 hours
14:42 < fenn> is there a sane way to make a timer for 6 hours?
14:45 < fenn> it also mentions using blue light as an inducer, so that seems like an easier solution
14:45 < yashgaroth> easier with DNA than mRNA, which is the usual delivery method these days, and even then it's not easy. You'd have one part of the plasmid expressing a low level of an inducer, and after ~a period of time~ there's enough of it to induce transcription of the inhibitor elsewhere on the plasmid
14:45 < yashgaroth> much easier to use light or something, even if you have to engineer a protein to be responsive to it
14:46 < kanzure> are we talking embryo or somatic adult?
14:46 < fenn> this is just crispr in general
14:50 < fenn> this paper uses inactive Cas9 to target specific regions for methylation as a way to do genomic imprinting in oocytes https://www.pnas.org/doi/pdf/10.1073/pnas.2115248119
14:50 < fenn> i'm sorta confused because wouldn't oocytes be haploid?
14:56 < fenn> it's a very brute force way to do it. "methylate here" and "demethylate here" at 7 different imprinting control regions
14:57 < yashgaroth> "two paternally methylated ICRs, the H19 and Gtl2 ICRs, form the major barrier that prevents bimaternal embryos from full-term development" so they had to methylated regions that are unmethylated in eggs, I guess it didn't matter that both copies were methylated since it seems they did it before diploidization
14:58 < yashgaroth> had to methylate*
14:59 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=5e86e2b1 Bryan Bishop: transcript: makepeoplebetter longevity >> http://diyhpl.us/diyhpluswiki/transcripts/makepeoplebetter/episode-002-longevity/
14:59 < yashgaroth> and yes it is very brute force, the in vitro gametogenesis people are/should be investigating it since gametes derived from iPSCs will have no methylation
15:21 < hprmbridge> kanzure> anybody have a canned response for her ? https://twitter.com/samira_kiani1/status/1635402884610412544
15:26 < L29Ah> aligned with who? :]
15:31 < L29Ah> how we can convince to accelerate things when we don't yet solved alignment problems among humans? accelerate panarchy first so these could be solved/compartmentalized better!
15:52 < hprmbridge> kanzure> not that tweet the other one
16:05 < docl> anyone else get llama.cpp working? https://github.com/ggerganov/llama.cpp
16:12 < superkuh> Yep. Running 30B now. It's nice they merged the interactive mode.
16:12 < superkuh> I also modified mine to dump the output to a fifo for my perl irc bots to communicate with. But it's... only working intermittently for some reason.
16:12 < L29Ah> the interactive mode kinda sux
16:12 < superkuh> And when you get to 2048 tokens long, that's end. No rolling buffer.
16:12 < L29Ah> since it seems to grab the context, and often it goes astray
16:13 < L29Ah> and not having https://github.com/ggerganov/llama.cpp/issues/91 is annoying
16:13 < L29Ah> can be hacked in an hour tho it seems
16:14 < L29Ah> but i don't see much use in the 13B model and can't run anything bigger
16:15 < superkuh> I played with the "fine tuned" stanford alpaca (7B lllama) but it wouldn't do "emergent" things like rhyme properly. The 30B seems like it wold be an entertaining bot but it doesn't impress me with super-human feats.
16:49 < kanzure> yesterday i should have typed "masked shoggoth thing" instead. whoops.
16:51 < kanzure> "Dariia Dantseva - Started Ukraine's First Biohacking Space" don't remember this name
16:53 < kanzure> https://github.com/tatsu-lab/stanford_alpaca https://twitter.com/abacaj/status/1635359347378376706 "cost less than $100 in cloud compute to fine tune"
17:01 < kanzure> ok so if "building new infrastructure is now illegal" https://twitter.com/pmarca/status/1635408070909243393 then i guess it might be implausible to expect people to be able to build more advanced things
17:21 < muurkha> https://nitter.fdn.fr/pmarca/status/1635408070909243393
17:22  * L29Ah recommends an url rewriter browser plugin for nitterizing
17:22 < muurkha> nmz787: why would you prioritize an additional vehicle over the lab then?
17:23 < L29Ah> lab vehicle
17:23 < muurkha> nmz787: is a macerating waste pump the same thing as a garbage disposal?
17:24 < muurkha> is there a particular rewriter plugin you'd recommend for Firefox?
17:24 < L29Ah> no, i don't use firefox
17:33 < kanzure> "For millions of years our living standards were extremely low-- extreme total poverty. We have barely begun to pick ourselves up out of the mud. We absolutely must accelerate."
17:37 < L29Ah> "ok."
17:39 < kanzure> yeah that's the expected reply
19:16 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0:854e:cd7c:7359:37f2] has quit [Quit: Leaving]
19:19 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has quit [Read error: Connection reset by peer]
19:56 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has quit [Ping timeout: 255 seconds]
20:57 -!- BEXCHA is now known as RDNCBRD
21:38 -!- RDNCBRD is now known as Mabel
22:21 < lsneff> You know, I might actually finish this thesis this week
22:32 < jrayhawk> lsneff is our new gradstudentbot
22:34 < lsneff> this is tru
22:50 -!- Mabel is now known as BEXCHA
22:57 -!- codaraxis__ [~codaraxis@user/codaraxis] has joined #hplusroadmap
23:00 -!- codaraxis [~codaraxis@user/codaraxis] has quit [Ping timeout: 250 seconds]
--- Log closed Tue Mar 14 00:00:10 2023