Aditya Sengupta
banner
adiastra.bsky.social
Aditya Sengupta
@adiastra.bsky.social
Astronomy PhD student at UCSC
Oh this is all so weird
November 8, 2025 at 7:35 AM
my one thought is would there be any tricks to get a faster inversion? I bet a lot of conventional things (ode solving for example) would be oom faster if that were the case
November 8, 2025 at 7:25 AM
AGI is hidden inside the commutative matmul 😔
November 8, 2025 at 7:13 AM
Also matrix multiplication is intensely symmetric imo, just not in the literal mechanics of how you'd do it by hand
November 8, 2025 at 7:12 AM
This is such an incoherent take? AI compute is bad because matmuls aren't commutative? If they were, would it be worth it? I don't think this even has its own internal logic
November 8, 2025 at 7:09 AM
pref for weekend or nah? could do as early as this Monday evening if not
November 8, 2025 at 5:52 AM
yes! i can't this weekend but down any time after
November 8, 2025 at 5:49 AM
Wait what happened to cause people to think there's a vibe shift
November 8, 2025 at 2:01 AM
oh I also got better at this over time, but not linearly - probably had my most work centric time ever this past summer
November 6, 2025 at 4:26 AM
(I just did a day of my class on this which is why I wanted to wait to formulate thoughts!)
November 6, 2025 at 4:22 AM
I've ~naturally found I need to build in recharge time for myself if I'm going to be effective at the difficult thinking tasks. So in a way it's the opposite of not having time for hobbies: I don't have the time *not* to do other things, even if the only reason to do them is to be better at science
November 6, 2025 at 4:14 AM
one of the biggest undergrad to grad shifts for me has been I get fewer useful hours in a day even if I'm nominally working the same amount, and I attribute that to doing more difficult things that need deeper thinking
November 6, 2025 at 4:04 AM
small group of course, but it's at least true to my anecdotal experience! So what I take away from this is - you don't have to put impossible productivity standards on yourself to be successful. Imo not only does it make you miserable, it doesn't actually work
November 6, 2025 at 4:02 AM
things take 1.7x as many hours as they think they will and 53% of tasks are completed in the planned week, but the thing I found the most interesting is more senior researchers aren't really better at estimating what they'll get done
November 6, 2025 at 3:58 AM
so there's this great April Fools paper by @kaleybrauer.bsky.social arxiv.org/abs/2103.16574 that I just used in my welcome-to-grad-school class, where a group of them tracked whether they finished all the tasks on their todo list each week for several months
"I'll Finish It This Week" And Other Lies
A small group of postdocs, graduate students, and undergraduates inadvertently formed a longitudinal study contrasting expected productivity levels with actual productivity levels. Over the last nine ...
arxiv.org
November 6, 2025 at 3:55 AM
this is real reminder to myself to come back to it with Thoughts
November 4, 2025 at 3:44 PM
You could translate it easily yeah but I don't think it was as intuitive pre digital clocks
November 1, 2025 at 8:47 PM
I'd bet it's based on analog clocks? You look at the hour hand, it's almost at 2, then you read the minute hand counterclockwise and subtract
November 1, 2025 at 8:34 PM
ngl your bike posting helps me here as someone who wants to live in LA but for the car dependence + who loves biking
October 29, 2025 at 4:54 AM
yeah the third point is why I haven't been able to share his work in relevant discussions - I like the conclusions if true, I just can't rely on the chain of logic that gets there
October 29, 2025 at 4:16 AM
I was like this as well! I'm so glad we're able to grow past being 14
October 29, 2025 at 1:54 AM