--- Log opened Fri Jun 28 00:00:37 2024
01:44 -!- Gooberpatrol66 [~Gooberpat@user/gooberpatrol66] has quit [Ping timeout: 256 seconds]
01:51 -!- Gooberpatrol66 [~Gooberpat@user/gooberpatrol66] has joined #hplusroadmap
02:26 < fenn> dear lazyweb, plz make a firefox addon that shows an image of the chemical structure in a standard orientation based on the main backbone when you mouse over a chemical name
02:57 < L29Ah> is lazyweb a brand of LLMs?
02:58 < fenn> no, i've spent the past half hour trying to get claude to do it
02:58 < fenn> so far i've got chemical names showing a blank box on mouseover, but for some reason the smilesdrawer js isn't drawing the structure
03:02 < L29Ah> > Although touted as the gold standard in minimizing the appearance of fine lines, wrinkles, enlarged pores and even acne, safety regulations have recently been put into effect to limit retinol’s concentration in skincare products. In April 2024, the European Union Commission announced that over-the-counter face creams, serums and gels could only contain up to 0.3% of retinol and 0.05% for body lotio
03:02 < L29Ah> ns and serums.
03:09 < nsh> fenn, try websim lazyweb
03:19 -!- TMM [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
03:19 -!- TMM [hp@amanda.tmm.cx] has joined #hplusroadmap
03:37 < fenn> websim wants me to sell my soul to google
03:37 < fenn> it's just claude anyway
03:38 < fenn> and also websim doesn't work at all for me 90% of the time, based on the links you share
04:34 < fenn> cool it worked! https://fennetic.net/irc/Chemical%20Structure%20Hover-0.11.png
04:41 < hprmbridge> kanzure> do you highlight the text first?
04:41 < fenn> it's a mouseover tooltip like wikipedia
04:42 < fenn> i'm using a hardcoded dictionary of chemical names and SMILES structures but i can maybe extend that dynamically in the future
04:43 < fenn> https://fennetic.net/irc/Chemical Structure Hover-0.12.user.js (tampermonkey will try to install it, sorry i dunno)
04:44 < fenn> ah the stupid debug rectangle
04:46 < fenn> welp it's super ugly but that's about as much as i can handle for tonight
04:48 < hprmbridge> kanzure> highlight to prompt claude or chatgpt might be useful too, better than search
04:58 < fenn> developing with userscripts and LLMs is surprisingly easy
04:59 < fenn> maybe a more general way to do this is to have the LLM directly plunk code into a userscript and run it
05:15 < fenn> the reason this took 2 hours instead of half an hour is that the browser may have needed some time to set up the canvas, asynchronous event loops and task queues, blah blah blah. i have no idea really. adding setTimeout(100ms) fixed it (fix courtesy of claude, not me)
05:27 -!- gl00ten [~gl00ten@193.147.150.201] has quit [Ping timeout: 268 seconds]
11:09 < hprmbridge> Eli> I’ve always felt this would be a money making idea. I don’t know how many organic chemistry students there are every year, but if you have an phone app that took pictures of a chemical name or the drawing of the structure and translated it into the reverse, you would effectively have a market paying you consistently every semester
11:11 < hprmbridge> Eli> The most difficult part would be drawing => chemical name. I’m assuming that you would need to feed a lot of chemical structures to teach it
11:25 < hprmbridge> nmz787> isn't that what SMILES and/or IUPAC is for though?
11:26 < hprmbridge> nmz787> there's like some rule for the starting/nucleation site... then you work iteratively from that atom until you cover all atoms
11:34 < hprmbridge> Eli> What if someone draws by hand?
11:52 < hprmbridge> kanzure> there is open source software that converts from chemical drawing diagrams to SMILES. I don't remember the name of the tool but it worked for me.
11:52 < hprmbridge> kanzure> some sort of chemistry OCR tool
11:53 < hprmbridge> kanzure> an app around that is a cool idea.....
11:54 < L29Ah> imago
11:54 < hprmbridge> kanzure> yeah that one
12:25 -!- TMM [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
12:25 -!- TMM [hp@amanda.tmm.cx] has joined #hplusroadmap
12:50 < hprmbridge> Eli> The things I would fund if I was a billionaire…
12:58 < L29Ah> Linux-based mobile device platform for SaaS enslavement of humanity
14:16 -!- darsie [~darsie@84-112-12-36.cable.dynamic.surfer.at] has quit [Remote host closed the connection]
14:16 -!- darsie [~darsie@84-112-12-36.cable.dynamic.surfer.at] has joined #hplusroadmap
15:55 < pasky> for vim users https://x.com/xpasky/status/1806823452516475106
15:55 < pasky> .m https://x.com/xpasky/status/1806823452516475106
15:55 < AugustaAva> ​twitter: <xpasky> @VictorTaelin @_xjdr Chat with Claude about your vim buffers: https://github.com/pasky/claude.vim very early heads-up, feedback / contributions welcome  ␤  ␤ (still ways to go to fulfill my vision and be nicer for newcomers; but self hosted for days already and using it feels so surreal!!)
16:05 -!- darsie [~darsie@84-112-12-36.cable.dynamic.surfer.at] has quit [Ping timeout: 246 seconds]
17:59 -!- TMM [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
18:00 -!- TMM [hp@amanda.tmm.cx] has joined #hplusroadmap
18:38 < hprmbridge> kanzure> https://twitter.com/balajis/status/1806774672949961016
19:07 -!- abetusk [~abetusk@syn-024-059-059-088.res.spectrum.com] has joined #hplusroadmap
19:08 < abetusk> Well, here we are in the future
19:23 < hprmbridge> kanzure> hello abetusk
19:25 < abetusk> hey kanzure
20:39 < fenn> .t https://lifescience.opensource.epam.com/imago/index.html
20:39 < saxo> Imago OCR
21:19 < docl> ah, looking like borohydrides might have promise as the decomposition reaction is highly endothermic. my apparent mistake was thinking mixing them with water would help. also, various azides have pretty strong endothermic decomposition reactions. in both cases the gas product needs to be adsorbed or consumed in another reaction (which can't be too exothermic)
21:28 < docl> also needs to have low enough activation energy (though maybe that can be delivered mechanically or electrically)
21:44 < docl> I have not fully persuaded myself endothermic reactions are the best way to achieve medical suspended animation via perfusable microencapsulated coolers, but it's an interesting rabbit-hole
21:45 < docl> another line of thinking is conformational changes in long polymers
21:49 < fenn> fwiw boranes are neurotoxic
21:50 < docl> I'm specifically considering microencapsulation to widen the possibility space, as the strongest reactions tend to be the most toxic and we really need to push the limits on this
21:55 < fenn> how about phase change reactions that occur with a pressure change
21:55 < docl> say you have a bunch of polymers highly aligned in parallel, forming a rod. then you squish the ends towards each other, causing the polymers to form random conformations, until you have a disc of intertangled molecules. these could also begin as isotactic molecules with side phenol groups (like polystyrene), and they could rotate and become atactic (randomized alignment of the side groups)
21:56 < docl> (this would be classified as a phase change)
21:57 < docl> the squishing effect might be achieved by charging a structure at the ends electrically
21:58 < fenn> so the rotation is more degrees of freedom and that increases the heat capacity, right?
22:00 < fenn> a mechanical spring pushing the ends apart makes more sense to me. i'm imagining arrays of these heat storage things with the ends compressed by flat plates, the plates being stacked many times
22:01 < fenn> they're in evacuated or gas filled micro bubbles injected wherever you need to cool, then put the body in a hydraulic pressure vessel at just above the freezing point. compress suddenly to increase the heat capacity and lower the temperature
22:05 < docl> hmm. I think when the plates press together they go to a lower temperature as the atoms around them can impart momentum to their squigly shapes
22:06 < docl> of the polymer molecules, that is
22:08 < docl> also can go to the orientation of the phenol groups (or maybe bigger cyclical groups). anyway it's also kinda about absorbing the randomness (entropy) of the gas molecule momentum vectors. the squished form has a large amount of random info storage capacity
22:23 < docl> basically you want something that starts out very orderly but loses order and converts it to structural strain or separation of charges or something every time it wriggles because a molecule bumped against it
22:32 < docl> hmm, appropriately designed polyrotaxanes could perhaps absorb heat efficiently
22:36 < fenn> is there a way to mechanically lock the rotaxane so it can't rotate?
22:37 < hprmbridge> Eli> Seems people smarter than me have already started working on the molecule recognition problem and open sourced it.
22:37 < hprmbridge> Eli> Paper: https://arxiv.org/pdf/2205.14311
22:37 < hprmbridge> Eli> Demo: https://huggingface.co/spaces/yujieq/MolScribe
22:37 < hprmbridge> Eli>
22:37 < hprmbridge> Eli> I feel like it would recognize my formulas because I always used one of those special OChem stencils when dranwing my hexagons (because I'm an idiot who wastes time). I might try uploading one of my stenciled molecules and see if its correct.
22:37 < hprmbridge> Eli> https://cdn.discordapp.com/attachments/1064664282450628710/1256483747990081637/image.png?ex=6680ef2d&is=667f9dad&hm=d380c490fac5c440f5799acfd910717f8e148f9843df20b5a6d2715750b7bbbc&
22:40 < fenn> 76-93% accuracy isn't amazing
22:42 < hprmbridge> Eli> I need to dig into the paper. I think some of the issues had to do with the types of inputs they used. So, I think they were saying the Japanese papers had issues (I need to reread). I would suspect normal molecules seen in OChem would do better. I'll input some of my drawings and see what happens
22:43 < fenn> you could also try like, putting the image into Imago with increased contrast
22:43 < hprmbridge> Eli> And since they're going after hand-drawn next, it seems like they will improve. Since it's open sourced, it seems like you could wrap it in some JavaScript crap to work cross-platform on a desktop or something.
22:43 < fenn> yeah crap it all up, internet of crap
22:44 < fenn> 100GB docker
22:44 < fenn> it's the way of the world and only crazies complain
22:51 < hprmbridge> Eli> Discord is made with Electron, a JS wrapper.
22:56 -!- darsie [~darsie@84-112-12-36.cable.dynamic.surfer.at] has joined #hplusroadmap
23:15 < hprmbridge> Eli> Just passed in these hand drawn molecules and got correct SMILES. It's current state will be the worst its going to get. Seems like there is quite a bit of room for improvement still. https://cdn.discordapp.com/attachments/1064664282450628710/1256493236068028486/Screenshot_2024-06-29_at_1.08.16_AM.png?ex=6680f803&is=667fa683&hm=25a91f640a938d667a3ed1955d018843702cc3d8b3ddd1cd3114f6d60815a3c6&
23:27 < hprmbridge> Eli> Tested some more. Hand drawn seems to have no problem as long as you have good writing. Also, you have to have clear drawing when something is pointing towards or away from you. But something like this image is no problem.
23:27 < hprmbridge> Eli>
23:27 < hprmbridge> Eli> Since it's open source, it seems like you could code it and then put it into an Electron app and see if Android/Apple will give you permission to take photos I guess? If it works, it would immediately be cross-platform, and the AI would only improve over time.
23:27 < hprmbridge> Eli> https://cdn.discordapp.com/attachments/1064664282450628710/1256496140367429652/Screenshot_2024-06-29_at_1.17.43_AM.png?ex=6680fab8&is=667fa938&hm=95aa6c57d765d29ff39da42c27951aae33582cc136908982dc8aa7d8f9d537d0&
23:41 < fenn> is electron really the easiest thing to use? i mean this literally is just some checkboxes and a button
23:42 < fenn> i can't fathom why people use electron, it's such a pig
23:44 < fenn> huh i didn't even realize that discord had a desktop client
23:46 < hprmbridge> Eli> It's probably the most stable out of the JS wrappers. There's JS stuff faster, but I don't really trust anything that hasn't had a few major apps built on it. The functional code can be written in Python or C++ with Electron as the wrapper. The functional code can theoretically be called via a Node module in Electron. If there's anything JS is good at, it's front-end UX and cross-platform
23:46 < hprmbridge> Eli> capability.
23:50 < fenn> but electron is 99% chrome, not JS
23:50 < fenn> like sure maybe the thing you as a developer touch is JS, but that's not what "it" is
23:51 < fenn> an entire OS for each "app"
23:52 < fenn> "an eight-week release cycle between major versions" sounds like hell
23:59 < hprmbridge> Eli> Wou;dn't the alternative be to code for each platform with Swift and Kotlin/C++? Then you have to have QA to ensure they look the same on all platforms. You run into the good, fast, cheap constraints. Just getting an MVP going cross-platform would be a fun project for a JS student. The Python code is already downloadable. So, you would need to see if you can get it working with Node with no issues
23:59 < hprmbridge> Eli> and then get camera permissions (I have no idea how locked down that is in phones)
--- Log closed Sat Jun 29 00:00:38 2024