samth.bsky.social
Associate Professor, IU Computer Science ·
Core Developer, @racketlang.bsky.social ·
Member, TC39 ·
Handler, Gravymaker ·
Bike Advocate, Bloomington IN
953 posts
1,543 followers
503 following
Getting Started
Active Commenter
comment in response to
post
competition makes salaries higher than they would be with a monopoly buyer. But that doesn't explain any particular level. And you could definitely hire a biology department at 60% of current salaries. It would just be much less good. And if everyone did that, the profession overall would be worse.
comment in response to
post
The interesting question is why professors/writers/etc get paid as much as they do, given that you could find people who would do it for less. (The answer is a combination of wanting better people and salary compression across fields )
comment in response to
post
It wrote the new thing I asked for in Python. And I am just estimating that it did it faster and with fewer failures than I would have expected in Racket.
I just told it to rewrite in Racket and that worked.
comment in response to
post
Hilariously, just after this, I asked codex to write something big using a long spec document, but in an existing Racket codebase, and it wrote it all in python (faster than it would have in Racket).
comment in response to
post
See bsky.app/profile/samt... for my experience in Racket.
I think Lean is a lot harder. Rust or Haskell probably in between.
comment in response to
post
Have you tried getting LLMs to write in more interesting languages than Python? I've been having good results with Racket.
comment in response to
post
The fun thing is you just say "write a bunch of tests" and then it does, and then subsequently it does better. But if you don't make it do that, it probably won't. Probably a lesson there.
comment in response to
post
12. I think I like expect testing.
Fin.
comment in response to
post
11. I don't think it's a fit for every kind of programming right now, but I write a lot of code that codex can do, and it's very obviously worth $20 a month right now.
comment in response to
post
10. Figuring out how to best give the model appropriate context without just giving it arbitrary internet access is tricky. I don't yet know the right answer. And oddly it regularly says that it tried to contact nytimes.com, although it's blocked.
comment in response to
post
9. The feature where it can try something 4 times, and you just pick the best/successful one, is quite effective. Quantity has a quality all its own.
comment in response to
post
8. You can do it effectively from your phone. You write a 1 sentence prompt, it does ~1 hour of coding in 10 minutes, then you write another sentence to tell it to fix whatever problem there still is.
comment in response to
post
7. It can do a lot of things in parallel, but beware merges. You can spawn almost arbitrarily many tasks, but if they have merge conflicts it's hard to resolve without just doing it yourself. I often ended up just giving up and redoing the conflicting change.
comment in response to
post
6. Tools are really important. That `raco fmt` exists is very necessary for getting good results, and it can just be completely transparent. Similar for testing tools, debugging tools, and so on.
comment in response to
post
5. It has somewhat less context than you'd expect. Even when you ask it to change something it wrote within a session, it spins up a whole new container. It's important to write the AGENTS file and give it access to the tools it needs.
comment in response to
post
4. It knows Racket. It knows how to write Racket programs, it knows scribble documentation, it knows Racket library APIs, it can use Racket idioms, etc. I'm sure the experience is better for Python and TypeScript but I didn't feel held back by using a minor language.
comment in response to
post
3. Tests are really important. The more tests you have, the more (a) you can trust the results and (b) the more likely it is to come back with results that are correct (because codex runs the tests). And you can have it write a lot of tests. Even the emacs integration has tests.
comment in response to
post
2. It really helps to know what you want. I was cloning an existing testing library and approach, I knew what libraries it should integrate with, and I knew how I wanted it to work. I don't think it would work as well for something I didn't know.
comment in response to
post
1. It really works. recspecs is almost 1500 lines of code, tests, and docs. It includes emacs integration, an extensive test suite, it works with rackunit (the standard Racket testing library), it has scribble docs. And I wrote effectively 0 lines of code.
comment in response to
post
Also, "Now Israel's getting tense, wants one in self defense. The Lord's our shepherd says the psalm, but just in case ... we better get a bomb"
comment in response to
post
I think "national brotherhood week" has a lot of relevance to today's multiracial Trump coalition.
comment in response to
post
Deep cut.
comment in response to
post
On the one hand obviously the civil war, reconstruction, and the end of slavery were not incremental reform. On the other hand they weren't the end of capitalism either.
comment in response to
post
Sure, it's hard and we keep failing. But it's less hard, and more likely, than a revolution in which we institute socialism/end the nation-state/abolish the US/you name it.
comment in response to
post
There are lots of things where it's easier to imagine a full scale revolution or collapse than major changes within the current framework but I think that mostly says something about how people imagine stuff.
comment in response to
post
comment in response to
post
Nice! Sadly since my parents moved I have not been back in a while.
comment in response to
post
Auburn is a really interesting study in right-wing YIMBYism. www.discoursemagazine.com/p/the-yimbye...
comment in response to
post
I replied over on twitter. I don't know if anyone is doing it here.
comment in response to
post
Several reporters are live-tweeting it, and I know some other people who are there.
comment in response to
post
This is a special case of "strengthening the induction hypothesis" which is often needed to prove a given property.
comment in response to
post
Right, that's what I mean. Politics is important, we wouldn't be on Bluesky if we didn't think that and that current Twitter's politics are bad, but I also want to discuss other stuff and there's not enough of that here.
comment in response to
post
But we're all here because of the politics -- very few people are on Bluesky unless they think new Twitter is bad.
comment in response to
post
A big issue is that bluesky is way more focused on politics than old twitter.
comment in response to
post
The issue is that they _talk_ in a way that is about winning elections. Eg, on Sunrise's home page they have:
REPORT: Sunrise 2024 General election Impact
comment in response to
post
But in fact that's not true, the mass mobilization approaches of the past are actually dead and the followers don't do what the leaders want and the leaders don't represent what the followers want.
comment in response to
post
So then people try to create that, fundamentally without success because 2025 (and because it's not something you can create). But everyone decides to talk in that mode anyway, and say that (a) their org represents lots of people who want something and (b) that it will lead to votes.
comment in response to
post
Then the center left transitions to this more think tank focused approach, and concurrently we lose a lot of elections. Then Theda Skocpol writes an article that basically says it works better when the change you want has a mass membership movement behind it.
comment in response to
post
It's about both at once. The stylized history is that we used to have these mass membership organizations that were transactional, so you could both see that lots of people cared about something and if you do what the leadership says the membership will vote for you.
comment in response to
post
I think unions are an in between case since they aren't DC organizations but do issue press releases about lots of issues. I think DSA is pretty obviously not one.
comment in response to
post
Gun control advocacy is genuinely something with broad support and local organizations the same as the Sierra Club. It's just not enough support.
comment in response to
post
For most of those, membership is just a term for giving money to them, though. The Sierra Club does have local groups that actually do things but the connection between that and national political positions is very attenuated.
comment in response to
post
Apparently that's how people refer to them in DC, as in "will the groups like this idea".
comment in response to
post
The distinction I'm making is between organizations like a union or DSA or the Women's Christian Temperance League where people have strong connections with the organization and organizations like NARAL or their counterparts on the right where people have an ideological but not social connection.
comment in response to
post
This isn't true on multiple levels. First, there are at least two issues that could be the "obvious" one you're referring to! Second, lots of people complain about The Groups being too dogmatic on abortion and housing and everything else!