parskatt.bsky.social
PhD student @ Linköping University
I like 3D vision and training neural networks.
Code: https://github.com/parskatt
Weights: https://github.com/Parskatt/storage/releases/tag/roma
1,331 posts
1,536 followers
382 following
Regular Contributor
Active Commenter
comment in response to
post
Angry at the moment (angry)
comment in response to
post
😬
comment in response to
post
🥺
comment in response to
post
Which is day after tmr over here ;)
comment in response to
post
Imo most coffee studies I think indicate that it's not bad for you, too difficult to disentangle all the reasons you might not be drinking coffee.
comment in response to
post
Yeah Ive started doing suffixes and I enjoy it.
comment in response to
post
I bet if you developed these types it made a lot of sense, but it doesn't explain anything to me as a user, and someone using my function definitely won't get it.
comment in response to
post
I don't like my code to be (similar to):
x: MyExtremelyVerboseAndVeryConfusingType[MyNiceType, SecondThingThatSomeoneProbablyUnderstands]
comment in response to
post
My main frustration is that overly verbose and precise types defined in e.g. numpy will force you to either adapt these extremely verbose types, or to in some way ignore the types.
comment in response to
post
If it works it works!
It maybe also depends on what you are typing and what type checker you are using. Pyright seems to work fine if I include enough ignore settings, but e.g. both pyrefly and ty fail.
comment in response to
post
Yep, not saying basic type checking is wrong. But if you start to include tensor shapes + datatypes it starts to go south.
comment in response to
post
Feels somewhat similar to hoping all mathematics will be done in lean. I really think there are benefits to having both more and less formal systems. In some cases you need formal guarantees, but not always. Python should not give any formal guarantees.
comment in response to
post
Not saying it's impossible to type correctly.
I'm saying that currently, trying to be perfectly correct with types seems to mostly create more problems than it actually solves.
comment in response to
post
I mean sure, in my case pyrefly can't even parse it.
I guess other typecheckers can.
comment in response to
post
It seems however that people try to use static type checking to ensure no bugs. I don't think the user cost is worth it.
comment in response to
post
Imo the main benefit of typing in Python is making your code easier to understand and reason about.
comment in response to
post
then*
comment in response to
post
If your types are more complicated than your code than probably typing is not making your code simpler.
comment in response to
post
Very sane and collected approach, instills confidence and hope for the future.
comment in response to
post
Gonna guess here: maps.app.goo.gl/E4y6C6MFDxVo...
Correct? I guess there are a few places on the complete other side of the earth that would fit as well ;)
comment in response to
post
I default to the garbage bin of my mind, which is a mishmash of many strange things, but it usually gets promply fixed by my coauthors 🙏
comment in response to
post
Physic
comment in response to
post
Could also be south...
comment in response to
post
Where's this? Looks like every swedish forest north of Linköping.
comment in response to
post
What are their policies on trash, transit, and housing?
comment in response to
post
Unfortunately neurips requires you to have at least 1 theorem in your paper, it is the law.
comment in response to
post
Don't let the math people read this, they will be upset.
comment in response to
post
Lemma: It doesn't matter that much that the proof is wrong, as the theorems are generally attempts to formalize intuitions about why something works. Since it already works there is no major issue.
comment in response to
post
There's also the neurips version: "the proof is in the appendix" (it's wrong)
comment in response to
post
The 8 hours part is not a joke.
comment in response to
post
Very grateful for this incredible machine. It's also a combined washer/drier. In just 8 hours you can get shrunken moist laundry!
comment in response to
post
Fake-it-instead-of-making-it?
comment in response to
post
There's no *only* boilerplate job, but there are many jobs where most of the tasks are boilerplate.
comment in response to
post
> automate majority
There's a big difference between majority and all.
I personally don't see any issue with majority, if you're talking 100% it's a different story, and seems very difficult to predict with certainty.
comment in response to
post
It seems to assume that past knowledge cannot be of any future use.
Most tasks we do have already been (partially) done before.
My hope is that this will mean humans will do more novel tasks in the future.
comment in response to
post
showyourstripes.info/c/europe/swe...
comment in response to
post
I'll let them know.
comment in response to
post
Persistence*...
comment in response to
post
Only if you have a talent for persistance.