--- Log opened Fri Mar 24 00:00:19 2023
00:59 -!- ANIMEX is now known as CRAVERDC
01:11 -!- codaraxis [~codaraxis@user/codaraxis] has joined #hplusroadmap
01:27 -!- codaraxis__ [~codaraxis@user/codaraxis] has joined #hplusroadmap
01:31 -!- codaraxis [~codaraxis@user/codaraxis] has quit [Ping timeout: 240 seconds]
01:35 -!- CRAVERDC [~Malvolio@idlerpg/player/Malvolio] has quit [Remote host closed the connection]
02:02 -!- CRAVERDC [~Malvolio@idlerpg/player/Malvolio] has joined #hplusroadmap
02:50 < hprmbridge> eleitl> Perry: thank you for confirming my point. This is why I'm not participating online anymore, apart from focused technical communities. So stick around, you won't see much of me here or anywhere else.
03:22 -!- CRAVERDC is now known as SELADOREII
04:02 < hprmbridge> Perry> who?
04:18 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0:a887:3042:909a:3e4d] has joined #hplusroadmap
04:21 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has joined #hplusroadmap
04:22 < hprmbridge> kanzure> zeloof
04:37 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
05:00 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has quit [Ping timeout: 246 seconds]
05:14 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has joined #hplusroadmap
07:30 < docl> I hope you guys will both stick around and that we can focus on technical matters as a community here. Perry has some expertise in the science of getting stuff done, which is a big emphasis here. and while I don't agree with doomsday scenarios which look a lot like catastrophizing to me, our narratives can vary. personally I don't want to die of old age or inadequate cryonics, for example. but I do 
07:30 < docl> think a growth mindset is healthier and more productive than focusing on that
08:16 < docl> my current project: get chatgpt to tell me about physics without resorting to greek characters
08:21 < hprmbridge> lachlan> Haha why?
08:23 < docl> because I don't have fluency with the greek characters and it's a persistent annoyance to see physics equations written like that. smacks of gatekeeping. at least, part of my mind thinks this. now that I'm seeing the forms side by side I am developing an appreciation for the compactness and standardization
08:24 < hprmbridge> lachlan> I see your point
08:24 < hprmbridge> lachlan> but I think they’re good
08:24 < hprmbridge> lachlan> it’s a small new language to learn though
08:25 < hprmbridge> lachlan> different glyphs often mean similar things, even in different contexts
08:25 < docl> but really I don't see how science literacy will progress if people keep using greek letters. they don't even mean the same things in all contexts
08:27 < hprmbridge> lachlan> the obvious replacement would just be using longer variable names, but that often makes equations harder to grok
08:28 < docl> yeah, there's no clear answer given how the brain / language learning / term comprehension cognition works. but maybe a dead simple expanded form side by side to standard glyphs would work best
08:31 < docl> using markdown convention for delimiters: |Time-independent Schrödinger equation|-ℏ²/(2m) ∇²ψ + Vψ = Eψ|(-reducedPlanckConstant^2 / (2 * particleMass) * laplacianOf(waveFunction) + potentialEnergy * waveFunction = totalEnergy * waveFunction)|
08:31 < hprmbridge> lachlan> God that’s horrific
08:31 < hprmbridge> lachlan> I’d be happy with a table defining each term
08:33 < hprmbridge> lachlan> like 𝛙 is always the wave function
08:33 < hprmbridge> lachlan> I think some of that got mangled by the bridge as well actually
08:33 < hprmbridge> lachlan> @kanzure did you write the bot?
08:34 < hprmbridge> lachlan> potential energy is often U, but not always
08:34 < L29Ah> greek is relatively fine
08:34 < docl> I mean, it's not like I can't look up psi to find out it's the wave function, but how am I supposed to understand what a wave function is until I've read and thought about it enough?
08:34 < L29Ah> perverted latin letters utilized by mathematicians are much worse
08:34 < hprmbridge> lachlan> You wouldn't, but you still wouldn't understand it if it was spelled "waveFunction" instead of '𝛙'
08:35 < hprmbridge> lachlan> ya, physics is at least somewhat grounded, math gets way wors
08:35 < docl> I'd have a mental hook to start thinking about it with
08:35 < hprmbridge> lachlan> they pull in stuff from semetic alphabets
08:35 < hprmbridge> lachlan> That's fair
08:36 < hprmbridge> lachlan> I had a thermo textbook use v in three different fonts to mean different things throughout the whole book
08:36 < L29Ah> also one can think of greek as garbled cyrillic ;)
08:36 < docl> yeah I've been using a similar approach on topology and linear algebra, group theory... I think the standard nomenclature will end up being what I actually use, but imo a logically valid camelCase works well as training wheels
08:37 < hprmbridge> lachlan> I'd be pretty happy with a button next to the equation that flipped between the compact and the verbose representations
08:37 < hprmbridge> lachlan> Would be very easy to do with papers that render to html
08:37 < docl> ooh nice thought
08:37 < L29Ah> yes, proper word-like identifiers would be great
08:39 < docl> ar5ive shows you html for papers there. https://ar5iv.labs.arxiv.org/html/1706.03762
08:39 < hprmbridge> Perry> You need to get a book like David J. Griffiths QM intro.
08:40 < hprmbridge> Perry> But to really understand that, you need linear algebra and differential equations under your belt. You can't avoid it. Math is the language of physics.
08:40 < docl> exactly. prerequisites can't be avoided
08:41 < hprmbridge> Perry> "wave function" is a really abstract notion. You probably want a book like French's "Vibrations and Waves" to get intuition about what wave functions mean.
08:41 < hprmbridge> Perry> There's no shortcuts. Understanding of this stuff is necessarily bottom up.
08:43 < hprmbridge> Perry> And if you don't get classical physics very well it's hard to move on to quantum. You even need things like the Lagrangian/Hamiltonian formulation of classical physics because all of QM is in those terms, there are no forces only potentials.
08:43 < kanzure> actually just a small app to do OCR on math equations and explain the terms or possible symbol meanings would be useful
08:43 < kanzure> you could also wire it up to openai's API and have it speculate about a few different options for what area of math/physics it is
08:43 < hprmbridge> Perry> And FWIW, the greek letters are pretty consistent inside a given discipline, and it's no worse than what happens when people use roman letters for certain quantities all the time. I'd rather read something in terms of alpha, pi, and h-bar than in terms of some less conventional notation.
08:44 < kanzure> just a small webpage that you paste an image into, and it does OCR and then explains it (or you tell it what area of math it is likely to be)
08:44 < hprmbridge> Perry> Kanzure: if you have read enough physics texts you don't need to speculate, reading the equations tells you immediately what kind of physics you're dealing with.
08:44 < kanzure> ain't nobody got time for that-- busy professionals in other disciplines need a cereal box decoder ring
08:44 < hprmbridge> lachlan> you just kind of get a feel for it
08:45 < hprmbridge> lachlan> I'm sure chatgpt could do it as well
08:45 < hprmbridge> Perry> Yes, but then chatgpt has the feel, not you.
08:45 < hprmbridge> Perry> If your goal is to *personally* understand something you can't avoid the work.
08:45 < kanzure> if you're a physics professional then yes you should know the physics symbols. but why shouldn't a lawyer be able to do a quick lookup and understand? why do they need to devote themselves to the guild of physicists first?
08:45 < hprmbridge> Perry> There are a ton of things AIs can help with (for example, finding compact symbolic or numerical solutions to equations) but it's not helpful for the thing to do my understanding for me if what I'm trying to do is learn it myself.
08:45 < hprmbridge> Perry> Lawyers will never understand physics. It's too hard.
08:46 < hprmbridge> Perry> I mean they can if they study the topic.
08:46 < docl> yeah I have an almost-but-not-quite notion of the hamiltonian. complex numbers, matrixes. I need fluency for it to gel. so I need to keep thinking and looking at examples
08:46 < hprmbridge> Perry> But you can't get an idea of what, say, the exchange interaction is without the math. It isn't even handwaveable.
08:47 < hprmbridge> Perry> "Why is it that you either have particles that can occupy the same quantum states like bosons or ones that can't like fermions?" "Because that's the set of available solutions to the equations when you exchange identical particles."
08:48 < hprmbridge> Perry> (That last bit isn't about the exchange interaction which explains part of chemical bonds. It's literally an artifact of the math and has no good handwave at all.)
08:49 < hprmbridge> Perry> Someday, we'll be able to directly exchange subnetworks from brain to brain to allow others to understand without work. For now, you have to build that stuff on your own. There's no royal road.
08:52 < docl> I don't disagree. however I distinguish somewhat between "fluent with the standard notation" and "understand the concepts". I think both are important/inevitable, but you can learn the concepts first (given a notation that works to do so)
08:54 < docl> and what happened in my case was developing a bit of a flinch reaction to standard glyphs. I suspect this is common for non-physicists / something physicists overcome or avoid early as students
08:56 < hprmbridge> Perry> Two hours with a greek alphabet printed on paper with the names of the glyphs and you will know them forever.
08:56 < hprmbridge> Perry> Seems like it's easy enough to do.
08:57 < hprmbridge> Perry> Just do the usual spaced repetition thing. Memorize them and then drill yourself at decreasing intervals afterwards.
08:57 < hprmbridge> jasoncrawford> Anki is good for that
08:57 < hprmbridge> lachlan> docl, what's your background?
08:57 < hprmbridge> lachlan> I need to set up anki, been procrastinating that sort of thing until i graduate
08:58 < docl> sure, translating it to a pronouncable word like "gamma" is easy. but fluency enough to know what a gamma matrix or gamma ray is is a different story
08:58 < hprmbridge> Perry> There are only a dozen that are really important anyway. pi, phi, alpha, and you've dealt with almost everything important in QM. h-bar isn't even a greek letter, it's an invented glyph.
08:58 < hprmbridge> Perry> It wouldn't help if it said "gamma" on the paper though instead of the letter.
09:08 < hprmbridge> kanzure> at least you could search on a search engine
09:08 < hprmbridge> Perry> You can just go to wikipedia.
09:08 < hprmbridge> Perry> It explains what alpha is, what phi is, and all the rest in this context.
09:09 < hprmbridge> Perry> You can even feed it the exact glyph.
09:09 < docl> yeah without knowing it's a matrix describing spin you have more work to do if it says "gamma".
09:11 < hprmbridge> Perry> https://twitter.com/perrymetzger/status/1639297315264516099
09:19 < docl> man, getting it to run proofs and correct itself would be sweet
09:37 < docl> lachlan: I'm a generalist, so it's kind of all over the place. some programming, lots of physics speculation practice from my dad (who did not grok SR or QM, but had a good grasp on electronics and classical physics). I haven't taken a physics class apart from a few youtube resources like MIT OpenCourseWare. same for linear algebra. I did take differential calculus and general chemistry in college, and 
09:37 < docl> have a computer science associate's (I was a college dropout, but went back to school in my 30's). since then I've been working as a software developer for a retailer
09:41 < docl> so I'm coming to a lot of this a bit late in life, almost 40. a lot of resources today that weren't around when I was younger, even just having unicode for all the standard math symbols is a big improvement
10:09 -!- cthlolo [~lorogue@77.33.23.154.dhcp.fibianet.dk] has joined #hplusroadmap
10:11 < hprmbridge> Perry> I find that learning math is something I do best with pencil and paper, even now.
10:51 < hprmbridge> kanzure> sir earl grey seems to have an interest in lipofuscin https://mobile.twitter.com/aubreydegrey/status/1639322883330830336
10:53 < hprmbridge> Perry> He's not wrong.
11:13 < hprmbridge> kanzure> indeed
11:27 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0:a887:3042:909a:3e4d] has quit [Read error: Connection reset by peer]
11:27 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0:800e:a59d:dd4e:d152] has joined #hplusroadmap
11:39 < L29Ah> alpaca seems to be out in the wild as well: magnet:?xt=urn:btih:e5322ab4676e24632a907fd9846234bb40265c4f&dn=ggml-q4
11:39 < L29Ah> should be in a ready for llama.cpp use format
11:45 < superkuh> You can talk to my wrapper for llama 13B over in (the terribly named) ##OpenAI . It's "llamar".
11:45 < superkuh> Er, alpaca 13B.
11:45 < superkuh> alpaca is a lot more likely to go off the rails and start regurgitating it's fine tuning prompts than llama (... with no fine tuning).
11:49 < L29Ah> superkuh: are you using the 4bit one through llama.cpp as well?
11:50 < superkuh> Pretty much. I fork it and wrap the interactive mode stdout/stdin. But also I added a lot much fflush.stdout calls to it more often and removed commented out all the debug info in the .cpp
11:51 < superkuh> The huggingface model weights python stuff is just too dependency hell/heavy.
12:30 < hprmbridge> kanzure> containers or virtualenv should have solved that.
13:07 -!- cthlolo [~lorogue@77.33.23.154.dhcp.fibianet.dk] has quit [Read error: Connection reset by peer]
13:14 -!- SELADOREII is now known as MARVEST
13:50 < L29Ah> sport competitions with allowed doping when?
15:44 < fenn> Perry> Someday, we'll be able to directly exchange subnetworks from brain to brain to allow others to understand without work.
15:46 < fenn> ^ i wasn't sure if this would ever actually become possible, but recently i've been playing around with "low rank adaptation" (LoRa) subnetworks with stable diffusion. you train a small network alongside your foundation model on a new concept or domain, and then it can be shared with other people as a small file (a few megabytes) and you can even load multiple of them at once, to combine
15:47 < fenn> concepts. it's amazing that just mashing them all together actually works, and it works well
15:47 < fenn> it works for both text and visual features
15:50 < L29Ah> but LoRa is a wireless modulation/protocol
15:51 < fenn> excuse me, LoRA
15:51 < fenn> happy? :)
15:57 < L29Ah> yes!
16:23 < L29Ah> indeed alpaca 13B is more cooperative, especially with chat-with-bob.txt prompt
17:01 < L29Ah> https://en.wikipedia.org/wiki/Khecarī_mudrā not sure if transhumanism
17:12 < hprmbridge> Perry> Unicode doesn’t make it through the bridge properly.
17:13 < hprmbridge> lachlan> Who’s running it?
17:23 < docl> kanzure: bot is showing garbage on the discord side for greek letter unicode chars like Ψ (should be psi)
17:24 < hprmbridge> Perry> It’s probably just not translating unicode correctly.
17:24 < hprmbridge> Perry> there might be a different encoding scheme on each side or something thing.
17:28 < hprmbridge> docl> Testing from the discord side:
17:28 < hprmbridge> docl> |Time-independent Schrödinger equation|-ℏ²/(2m) ∇²ψ + Vψ = Eψ|(-reducedPlanckConstant^2 / (2 * particleMass) * laplacianOf(waveFunction) + potentialEnergy * waveFunction = totalEnergy * waveFunction)|
17:41 -!- sgiath [~sgiath@mail.sgiath.dev] has quit [Ping timeout: 240 seconds]
17:41 -!- sgiath [~sgiath@2a02:25b0:aaaa:aaaa:a3c3:ed4b:6b06:0] has joined #hplusroadmap
17:44 < kanzure> https://www.forbes.com/sites/alexknapp/2023/03/17/why-billionaires-ken-griffin-and-eric-schmidt-are-spending-50-million-on-a-new-kind-of-scientific-research/
17:44 < kanzure> docl: the bot was written very poorly; it was just something i found on github. find me something better?
17:45 < kanzure> Perry: for brain-to-brain transfer consider https://www.sciencedirect.com/science/article/pii/S2589004219303839
17:51 < kanzure> https://www.intel.com/content/www/us/en/newsroom/news/gordon-moore-obituary.html
18:18 -!- MARVEST is now known as CAVERNOUS
18:23 -!- sgiath [~sgiath@2a02:25b0:aaaa:aaaa:a3c3:ed4b:6b06:0] has quit [Ping timeout: 265 seconds]
18:23 -!- sgiath [~sgiath@2a02:25b0:aaaa:aaaa:a3c3:ed4b:6b06:0] has joined #hplusroadmap
18:35 < hprmbridge> lachlan> managed to do it: https://twitter.com/_nymx_/status/1639438319325962242
18:38 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0:800e:a59d:dd4e:d152] has quit [Quit: Leaving]
18:40 < hprmbridge> Perry> does it still work after you’ve decimated it?
18:40 < hprmbridge> Perry> How much worse?
18:40 < hprmbridge> lachlan> indeed it does
18:40 < hprmbridge> lachlan> I haven’t tested it a huge amount though
18:40 < hprmbridge> lachlan> falls into mode collapse more quickly it seems
18:41 < hprmbridge> lachlan> I wanted to get it working with llama.cpp, because the sampler and whatnot is good, but it was too much work
18:52 < hprmbridge> Perry> https://twitter.com/perrymetzger/status/1639444623322935297
18:54 < hprmbridge> kanzure> the quest for consciousness set us back by 100s of years.
18:54 < hprmbridge> lachlan> 100s?
19:43 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has quit [Ping timeout: 240 seconds]
22:18 -!- cthlolo [~lorogue@77.33.23.154.dhcp.fibianet.dk] has joined #hplusroadmap
23:03 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has quit [Ping timeout: 252 seconds]
23:18 -!- CAVERNOUS is now known as HAVAGOAT
--- Log closed Sat Mar 25 00:00:20 2023