--- Log opened Wed Dec 25 00:00:28 2024
05:07 -!- darsie [~darsie@84-112-12-36.cable.dynamic.surfer.at] has joined #hplusroadmap
07:11 < nsh> semantic search of arxiv papers: https://papermatch.mitanshu.tech/
07:30 < L29Ah> told it "Longevity HIIT", the first result has no mentions of HIIT anywhere
07:46 < nsh> --
07:46 < nsh> I used MixedBread's [^1] embedding model to generate vectors from the abstracts. I store and search similar vectors using Milvus [^2] and finally use Gradio [^3] to serve the frontend. I update the vector database weekly by pulling the metadata dataset from Kaggle [^4].
07:46 < nsh> To speed up the search process on my free oracle instance, I binarise the embeddings and use Hamming distance as a metric.
07:46 < nsh> -- https://news.ycombinator.com/item?id=42507116
07:49 < nsh> not getting exactly what was asked for is part and parcel of how 'semantic' search differs from [ranked] matching search 
07:49 < nsh> the two approaches can however be complementary
07:51 < nsh> the acronym HIIT quite plausibly wasn't in the MixedBread training data
07:51 < nsh> 'The model was trained on a vast dataset of over 700 million pairs using contrastive training and fine-tuned on more than 30 million high-quality triplets using the AnglE loss function. This extensive training enables the model to adapt to a wide range of topics and domains, making it suitable for various real-world applications and Retrieval-Augmented Generation (RAG) use cases.'
08:17 -!- TMM [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
08:17 -!- TMM [hp@amanda.tmm.cx] has joined #hplusroadmap
10:09 -!- Hooloovoo [~Hooloovoo@hax0rbana.org] has quit [Ping timeout: 244 seconds]
10:11 < docl> https://www.youtube.com/watch?v=b0cU-CmjKNY // Seven Science Carols | A Capella Science
11:04 -!- Hooloovoo [~Hooloovoo@hax0rbana.org] has joined #hplusroadmap
11:17 -!- Hooloovoo [~Hooloovoo@hax0rbana.org] has quit [Ping timeout: 272 seconds]
11:32 < fenn> .wik Russell Brain, 1st Baron Brain
11:32 < saxo> "Walter Russell Brain, 1st Baron Brain (23 October 1895 – 29 December 1966) was a British neurologist. He was principal author of the standard work on neurology, Brain's Diseases of the Nervous System, and longtime editor of the neurological medical journal titled Brain." - https://en.wikipedia.org/wiki/Russell_Brain,_1st_Baron_Brain
11:37 -!- ike8 [e8f913dbdf@irc.cheogram.com] has quit [Remote host closed the connection]
11:37 < hprmbridge> kanzure> https://eon.systems/
11:38 -!- ike8 [e8f913dbdf@irc.cheogram.com] has joined #hplusroadmap
11:40 -!- Hooloovoo [~Hooloovoo@216.169.5.238] has joined #hplusroadmap
12:09 -!- Hooloovoo [~Hooloovoo@216.169.5.238] has quit [Ping timeout: 265 seconds]
12:17 < fenn> michael andregg finally found a use for those millions of electron microscopes on a chip?
12:18 < fenn> has eon subsumed nectome?
12:18 < hprmbridge> kanzure> unclear to me. asking russell.
12:25 < fenn> i don't remember if i've talked about it here much, but in AI there's this technique called knowledge distillation where you can train one network to predict the activations of another network, and then you get the same behavior out the end without having to train from scratch
12:26 < hprmbridge> kanzure> search the life we have had this conversation before
12:26 < hprmbridge> kanzure> er, the logs
12:27 < fenn> it seems a lot easier to hook up a relatively small number of transcievers to neurons in a brain and predict those activations, vs completely scanning every synapse and dendrite and axon and cell body and so on, then somehow come up with a way to turn that structural data into functional data
12:28 < fenn> just collect functional data to start with
12:28 < fenn> of course this requires the brain to be operating, and could take a really long time to cover a large fraction of state space
12:30 < fenn> distilling also completely bypasses the rather difficult technical problem of emulation. since the student network can be any architecture at all, you can train a model that matches your inference hardware really well
12:33 < fenn> stimulating the brain to cover such a wide variety of states would itself be a subjectivity-fracturing experience, you wouldn't know if you were the clone or the original
12:35 < fenn> uh, unless you've been thusly-stimulated and then cloned, and could tell the difference (presumably being stimulated twice would be a similar experience, but being cloned would be different from being stimulated)
12:38 < fenn> you might not remember the process either way
12:53 < hprmbridge> kanzure> wasn't this a greg egan story
13:08 -!- Hooloovoo [~Hooloovoo@hax0rbana.org] has joined #hplusroadmap
13:20 < fenn> yes, permutation city. but then it goes off the rails with some "but everything mathematically possible is real!" crap at the end
13:21 < fenn> also the movie "abre los ojos"
13:21 < fenn> i do not recommend jumping off of tall buildings
13:22 < hprmbridge> kanzure> no the other story
13:22 < hprmbridge> kanzure> about the diamond that confuses itself
13:22 < fenn> i must not have read it
13:23 < fenn> oh it's literally called "Transition Dreams" heh
13:24 < hprmbridge> kanzure> "learning to be me"
13:25 < fenn> 'dreams that your new, robotic brain has as it is being "filled up" with the patterns copied from your old, organic brain'
13:26 < hprmbridge> kanzure> not a diamond, a jewel, sorry
13:26 < fenn> battle angel alita has a similar thing
13:27 < fenn> worth noting that the new architecture won't necessarily learn in the same way as the old one
14:21 < hprmbridge> kanzure> nectome is involved, somehow
14:29 -!- Hooloovoo [~Hooloovoo@hax0rbana.org] has quit [Ping timeout: 246 seconds]
14:31 -!- Hooloovoo [~Hooloovoo@hax0rbana.org] has joined #hplusroadmap
14:40 -!- Hooloovoo [~Hooloovoo@hax0rbana.org] has quit [Ping timeout: 265 seconds]
14:42 -!- Hooloovoo [~Hooloovoo@hax0rbana.org] has joined #hplusroadmap
14:52 -!- Hooloovoo [~Hooloovoo@hax0rbana.org] has quit [Ping timeout: 276 seconds]
14:57 -!- Hooloovoo [~Hooloovoo@hax0rbana.org] has joined #hplusroadmap
16:55 -!- Hooloovoo [~Hooloovoo@hax0rbana.org] has quit [Ping timeout: 252 seconds]
16:56 -!- Hooloovoo [~Hooloovoo@hax0rbana.org] has joined #hplusroadmap
17:06 -!- darsie [~darsie@84-112-12-36.cable.dynamic.surfer.at] has quit [Ping timeout: 244 seconds]
17:12 -!- Hooloovoo [~Hooloovoo@hax0rbana.org] has quit [Ping timeout: 244 seconds]
17:13 -!- Hooloovoo [~Hooloovoo@hax0rbana.org] has joined #hplusroadmap
17:16 < L29Ah> https://neuromatch.social/@elduvelle/113715250479085716
17:16 < L29Ah> .t
17:16 < saxo> El Duvelle: '@jonny Yes! Incredible!!Here's a video from the…' - Neuromatch Social
17:17 < L29Ah> ants problem-solving video
20:23 -!- mxz__ [~mxz@user/mxz] has joined #hplusroadmap
20:24 -!- mxz [~mxz@user/mxz] has quit [Ping timeout: 276 seconds]
20:24 -!- mxz_ [~mxz@user/mxz] has quit [Ping timeout: 245 seconds]
20:24 -!- mxz__ is now known as mxz
20:29 -!- TMM [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
20:29 -!- TMM [hp@amanda.tmm.cx] has joined #hplusroadmap
21:37 -!- gl00ten [~gl00ten@220-138-192-182.dynamic-ip.hinet.net] has joined #hplusroadmap
21:41 -!- gl00ten2 [~gl00ten@101.12.98.149] has joined #hplusroadmap
21:43 -!- gl00ten [~gl00ten@220-138-192-182.dynamic-ip.hinet.net] has quit [Ping timeout: 260 seconds]
22:18 -!- gl00ten2 [~gl00ten@101.12.98.149] has quit [Ping timeout: 264 seconds]
22:30 -!- mxz_ [~mxz@user/mxz] has joined #hplusroadmap
23:19 < hprmbridge> poppingtonic> hey all
23:22 < hprmbridge> poppingtonic> trying out msep.one since it came out, has anyone else played with it?
23:23 < hprmbridge> poppingtonic> FYI: https://github.com/MSEP-one/msep.one
--- Log closed Thu Dec 26 00:00:29 2024