if anyone is willing to organise this in the himalayas this summer we would be forever grateful to you and just imagine the service you would be doing to everyone being there!
it is time to go back to the original ways
"and when people couldn't find this oscillation for a few days....they resorted to releasing these models on their own trying to experience the lost bliss"
> new model comes
> people get happy/sad with it
> new model goes
> people get happy/sad with it
> new model comes again.....
when will the people realise....? there are better ways to spend your time than this
> new model comes
> people get happy/sad with it
> new model goes
> people get happy/sad with it
> new model comes again.....
when will the people realise....? there are better ways to spend your time than this
also the ai bubble shit takes, please don't force me to leave this place, nothings bursting because it never existed in the first place, the only bubble is 🫵
woah would have loved if this released ~1 month ago coinciding with the @GRaM_org_ competition for airflow prediction, love the DEC approach they take (my very naive approach of building something of a TNO looked really bad)
very exciting work again!
We also design an anisotropic Darcy benchmark with randomly oriented face-wise diffusion tensors. Here, respecting the native topological degree of physical quantities matter and lead to more faithful operator learning, reflected in significant performance gains for our TNOs.👇
would be fun to try it out over the CC's from the dataset this weekend and see where it lands up on the leaderboard (hopefully they release the code until then and make this easier)
maybe it was all just about the 10 different tabs of paul graham essays that you have open on a thursday for light reading but now you want to spend rest of the day reading and writing about random things
was watching this super interesting video about cut locus of riemannian manifolds (i dont even remember how i stumbled upon it a week ago), the speaker started in english and just when i thought this is it he switched to french
anyways this is how i got back to learning french
pantheon is such a riveting watch, makes you think about so many things at the same time. the loop is something like
> hear about something x
> have insane opinions about x
> want to write about it
> almost instantly y drops into the picture
> cycle continues
New research from @japhba and I!
Activation Oracles are a pretty cool interpretability tool. They answer natural questions about activations, but they suffer from vagueness and hallucinations. Can AO training be improved?
Turns out: Yes! We identify four fixes that make AOs substantially more useful!
Short blog on why I think SSL isn't scaling.
tldr we're suppressing too much information from the SSL loss so that anything is learnt at all, but this limits the total set of things we may be able to learn.
jchencxh.github.io/blog/scal…
while on this topic i came across this very neat compilation by @_vaishnavh, ill leave you with this line anon!
"I contend that you cannot learn this emotional maturity efficiently through the brute-force of the research process alone. You need to read."
the value of understanding the history of your field is highly undervalued and you know what it probably is so interesting that once you start it becomes impossible to abandon it mid way
p.s its also coupled with multiple side effects that help on the journey!