Some of this might actually be fixed, or at least, in a position where it can be fixed. We had a problem for a while where updating the software was impossible because it caused crashes and largescale fixes were sketchy due to how badly tested it was. Thankfully, with the rise of AI, I can (and have) just point Claude at the codebase and say "write a ton of tests kthx" and now it's a lot less questionable to do work on.
It actually found a few bugs while writing the tests which was pretty funny.
They are undoubtably useful, especially in a domain like writing tests. If you are writing tests post-facto, you have a very well defined problem with a constrained domain, and a massive corpus to consult in terms of style, strategy, etc, etc. We live in a world where hundreds of millions (and I'm probably lowballing) of tests have been written, including tests for some of the hardest math and logic problems known to man. Every leet code website and programming competition out there uses a suite of test cases to verify correctness of submissions.
Honestly writing software tests is almost the optimal use case for an LLM.
I write lots of tests at work, and it's all incredibly domain-specific and trying to use LLMs for it just leads to laughably incorrect test expectations. Hence why I have trouble imagining it just working out of the box.
Claude is probably the best-integrated system out there; use the commandline tool because it's able to go research your codebase on its own.
Make it a CLAUDE.md with general architecture. It can make a CLAUDE.md for you with the /init command, but read over it by hand to make sure it's right.
Make sure thinking is enabled (hit "tab"). If you're asking it to do a really hard thing, use the "ultrathink" keyword, which will give it more space to think.
If you have reference tests to use as a model, point it at those. Telling it something like "read existing tests and use the same general style" can help a lot; if you can be more specific, do so.
Makes sense, I think it all comes down to the corpus of public information available. If similar work can be scraped from one of the half billion repos on github its probably great, otherwise it likely isn't. Most of my work is either web stuff, infra as code, or business logic governed by public policy (rather than internal rules) so it's pretty ideal.
I imagine it would be significantly worse if I were still in global finance, but we often had the same issue as the LLMs. Lack of information and often having to switch tasks while we wait a week for a request for some bespoke implementation details to go up the chain, over to IBM, and back.
This website is a place for people who want to move past shady thinking and test their ideas in a
court of people who don't all share the same biases. Our goal is to
optimize for light, not heat; this is a group effort, and all commentators are asked to do their part.
The weekly Culture War threads host the most
controversial topics and are the most visible aspect of The Motte. However, many other topics are
appropriate here. We encourage people to post anything related to science, politics, or philosophy;
if in doubt, post!
Check out The Vault for an archive of old quality posts.
You are encouraged to crosspost these elsewhere.
Why are you called The Motte?
A motte is a stone keep on a raised earthwork common in early medieval fortifications. More pertinently,
it's an element in a rhetorical move called a "Motte-and-Bailey",
originally identified by
philosopher Nicholas Shackel. It describes the tendency in discourse for people to move from a controversial
but high value claim to a defensible but less exciting one upon any resistance to the former. He likens
this to the medieval fortification, where a desirable land (the bailey) is abandoned when in danger for
the more easily defended motte. In Shackel's words, "The Motte represents the defensible but undesired
propositions to which one retreats when hard pressed."
On The Motte, always attempt to remain inside your defensible territory, even if you are not being pressed.
New post guidelines
If you're posting something that isn't related to the culture war, we encourage you to post a thread for it.
A submission statement is highly appreciated, but isn't necessary for text posts or links to largely-text posts
such as blogs or news articles; if we're unsure of the value of your post, we might remove it until you add a
submission statement. A submission statement is required for non-text sources (videos, podcasts, images).
Culture war posts go in the culture war thread; all links must either include a submission statement or
significant commentary. Bare links without those will be removed.
Jump in the discussion.
No email address required.
Notes -
It's not just you. Working on it. :)
(Should be better now, I'll let it sit for half an hour or so and see how it's going. More work may be needed.)
(Edit: Sigh, not solved. Still working on it.)
(This may now be fixed; I think we were just running on bad hardware.)
It's been kinda slow for a long, long time (intermittenly most of last year?).
Not a huge deal because once it loads you usually don't have a problem but sometimes it'd take 10seconds to render etc. Or so I remember.
Some of this might actually be fixed, or at least, in a position where it can be fixed. We had a problem for a while where updating the software was impossible because it caused crashes and largescale fixes were sketchy due to how badly tested it was. Thankfully, with the rise of AI, I can (and have) just point Claude at the codebase and say "write a ton of tests kthx" and now it's a lot less questionable to do work on.
It actually found a few bugs while writing the tests which was pretty funny.
Wow, that is truly amazing. And, hilariously, it's going to be a very powerful datapoint for our constant "are LLMs actually useful?" debates.
They are undoubtably useful, especially in a domain like writing tests. If you are writing tests post-facto, you have a very well defined problem with a constrained domain, and a massive corpus to consult in terms of style, strategy, etc, etc. We live in a world where hundreds of millions (and I'm probably lowballing) of tests have been written, including tests for some of the hardest math and logic problems known to man. Every leet code website and programming competition out there uses a suite of test cases to verify correctness of submissions.
Honestly writing software tests is almost the optimal use case for an LLM.
I write lots of tests at work, and it's all incredibly domain-specific and trying to use LLMs for it just leads to laughably incorrect test expectations. Hence why I have trouble imagining it just working out of the box.
I don't know what you've tried, but:
/initcommand, but read over it by hand to make sure it's right.More options
Context Copy link
Makes sense, I think it all comes down to the corpus of public information available. If similar work can be scraped from one of the half billion repos on github its probably great, otherwise it likely isn't. Most of my work is either web stuff, infra as code, or business logic governed by public policy (rather than internal rules) so it's pretty ideal.
I imagine it would be significantly worse if I were still in global finance, but we often had the same issue as the LLMs. Lack of information and often having to switch tasks while we wait a week for a request for some bespoke implementation details to go up the chain, over to IBM, and back.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link