CIRCLE WITH A DOT

@cross @shriramk @jfdm @csgordon @lindsey @jeremysiek yes, how stupid it feels to be literally using an extreme number of neurons to infer something that *has a semantics*. I keep telling this to people and the response I get is something like: well, that’s tied up in a ton of other maybe-useful stuff it inferred.

@cross @shriramk @jfdm @csgordon @lindsey @jeremysiek oh no, I think it totally will be for many contexts. Not all surely, and having a solid understanding of all code will still be important surely, but for increasingly fewer tasks I expect

@shriramk @jfdm @csgordon @lindsey @jeremysiek I think once you trust that the student could in principle write the code (and they're treating it like code the prof gave them, code their coworker wrote, etc.) then what you're saying is right. The concern is: "go through whole college career and just have claude code do every single homework assignment with very little intellectual effort." Of course, many would argue that this is a failure of the curriculum design--but it will inevitably take time to catch up.

@shriramk @jfdm @csgordon @lindsey @jeremysiek okay, wow--I did not really expect that. Interesting, I will have to think about that.

@va2lam @shriramk @jfdm @csgordon @lindsey @jeremysiek yeah, here our local thing is called "Resilient AI" (since that covers, you know, everything).

@jeremysiek fully-automated natural-language-based software engineering is here, in a big, big way. I think it has the potential to be massively powerful in the right circumstances, but a major shift happened around December I noticed.

@bool I hear you on this and mostly I agree with you, but I definitely think we're going to see a big exodus of applicants (we're already seeing this here in CS) when the AI can just do most of the assignments even driven by someone totally unqualified.

@jeremysiek mid-level ugrad PL class (interpreters, Boehm encoding.) in Racket--this is Claude Code (with access to the test suite) and Codex

Welp, for the first semester ever, SOTA LLMs can do *every single assignment, from scratch (readmes, etc.), and get 100%*. It is frustrating because it would now be quite tough to make it so that LLMs *couldn't* do this without also making it so that the assignments are obfuscated past the point of offering educational value...