Aditya Sengupta

Also without models that are more local/without the huge environmental footprint/without being locked to the big ai companies. I hope we see democratization of LLMs so we can make narrowly scoped tools that actually work/have guardrails and the hype can end

I think the aspiration of pro AI people is to push up the abstraction barrier so you don't have to know code specifics, in the way that many scientists now don't need to know low level computer architecture unlike a few decades prior. But you can't get there without reproducibility

Yep fully agree; I was imagining this at the grad student+ level. I think earlier on you'll be totally lost if you don't hand write it all yourself

aditya-sengupta.github.io/vibescience/
I also think you will learn a lot by writing some things by hand, but I could imagine that doesn't always apply. But for research integrity I think you do always have to do your vetting by hand no matter how you got the code

ok I have a take here: i don't inherently mind ai generated code if it moves the person's focus to writing tests/sanity checking, because that should be happening for human written code too and is arguably where the physics is. But you do have to actually do the checks, and lots of ai users don't

I have literally said things like this to my friends and I go on walks so I can completely understand things!

Our prospective students are flying out here today, I think the Northeast ones are just barely going to make it but who knows

As a grad student who's recently taught a class for first-year grads I really like the section on focusing on learning over just producing work! I think a grad costs ~as much as a postdoc, and is bound to produce less, but it's worth it for our learning - it's nice to see the field-wide view of that

yay!!!!!

it's basically this but you'll also get like an arrow pointing left/down saying we want to go here

I just got done with spirit of lyot and the number of times I saw either this or the hr8799 video, people were apologizing for still doing it by the end

I like reading blog posts or latex docs more personally! But would ultimately read whatever format

so fair, sorry he said that :(

The most charitable read I've got is "we had opportunities/advantages you didn't, so you should go easier on yourself" but even then it's really poorly expressed

The title and author list of the linked paper, "On-sky Demonstration of Second-stage Wave-front Control with a Photonic Lantern"

My paper demonstrating second-stage wavefront control using a photonic lantern at the Shane 3m telescope at Lick Observatory has been published in the Astronomical Journal! Good start to the year!
doi.org/10.3847/1538...
via @ioppublishing.bsky.social

The more these scenarios come up the more I realize my position is incoherent - i know there must be a concrete opinion that broadly agrees with my gut sense of "I don't trust it but maybe that's okay for some stuff" and captures all the nuances and fully guides how we should teach and learn and invent stuff, but I'm not there yet

Omg literally something I said the other day thinking about how to teach/do research now. I've felt so bad that I can't resolve these conflicts so this helps

Yes Minister invented this guy youtube.com/clip/UgkxZwz...

as someone who may not be a researcher if not for prof layton this is so sad
but also, not clear to me if this is more like "we should let ai write the layton puzzles" or "we should write the game in cursor instead of vscode" - even when specified to genai we seem to have terminology issues

Minute Cryptic - 25 December, 2025
"Christmas now?" (7,3)
🟣🟣🟣🟣🟣🟣🟣🟣🟣🟣🟣🟣
🏆 0 hints – 3 under the community par (38,328 solvers so far).
www.minutecryptic.com?utm_source=s...

holmes, columbo, poirot are all exactly this! feels like very normal for anything episodic/serialized

Fair enough but I also don't think knowing physics is a 0/1 thing! There may be (always is) more to learn but dividing up time into small intervals and looking at whether events occur in them + knowing that that approach can describe physical phenomena does count as knowing some part of physics

i think this totally counts as knowing physics tbh!

Indirect Detection

Q why would you xkcd.com/2071/ me like this

wow ignored

yeah I think that's my take on the whole album lmao

i fully forgot this album had a song called The World's First Ever Monster Truck Front Flip

look sometimes you like an album for no other reason than that it came out in your freshman year of college and that's *fine*

Six playing cards with curses written on them, each describing something seekers have to do in a game of hide and seek: (1) find a surfboard, wetsuit, or boat (2) walk your bike (3) lock your bike at a legitimate parking rack (4) have your gear chosen by the hider (5) visit a local chain business (6) ask a question from a street without lights

wrote some custom curses for jet lag hide and seek on bikes in Santa Cruz!

Honestly yeah I've made the US train experience work surprisingly well but there's almost never a coffee/sandwich option within a "can go and come back in time for the next train" radius

Wow look who's at the top of astro-ph.IM today arxiv.org/list/astro-p...