nathan
nds.bsky.social
nathan
@nds.bsky.social
ml for science
thank you cambridge university library for sharing, can't wait to try 👍
November 3, 2025 at 3:53 PM
"the value of art is how much you would risk to be in its presence" — I found this alone quite moving!
October 25, 2025 at 10:31 AM
yeah i think that's the camp i sit in! i will say that discord is kinda nice if you have a hybrid working setup — hopping in/out of audio calls has less friction, and you can have audio channels that function as break rooms that you can chill in :)
October 17, 2025 at 3:14 PM
It's so interesting... Scaling models & data definitely feels like our best path to model discovery, but somehow feels like just a brute-force way to get around not setting the right constraints on the learning problem.

That being said, we have no better methods for this kind of performance!!
September 25, 2025 at 11:03 AM
oh this is cool, thanks!! my paste gives

^-- SC2164 (warning): Use 'cd ... || exit' or 'cd ... || return' in case cd fails.

it already caught me 🤫
September 15, 2025 at 4:54 PM
congrats :) sounds awesome!!
September 4, 2025 at 1:16 PM
This looks like a great lineup!! I'd like to be around but this is me just discovering the event 😅
September 2, 2025 at 4:30 PM
yeah i'd love that! you have any mornings coming up with a free hour? send through your email in whatever way you'd like and I can make a meeting invite :)
July 22, 2025 at 10:03 AM
i should go through your pixi workflow at some point — ive hard-swapped to pure uv with lockfiles for a recent project and that’s been great!
July 21, 2025 at 9:28 AM
well, maybe not *most* frustrating; i'd prefer if it wasn't cheaper to literally fly between cities in the uk due to how expensive rail travel is...
May 18, 2025 at 11:37 AM
taking de-bugging to a whole new level 😌
May 8, 2025 at 8:06 AM