Helga19h ago

I start reading what is supposedly a powerful personal story about burnout in open source. Then I start to see the pattern - that familiar tempo. Admittedly there's no "Not x. Not y. Just z'. But there's that faint whiff, like the neighbor is burning toast. The paragraphs are just a bit too regular. Something about the headings. Yeah my dude, you used ChatGPT to write this, didn't you. I'm sure of it. Not sure enough of just how much to comment on the post, and I don't have the energy to get in an argument about it.

Ai both degrades trust - making us suspicious of everything we see and read - and sucks the life and personality out of absolutely everything it touches. I'm so sick of starting to read and going, oh, nope this is machine-generated nothingness. I hate this. I hate all of it.

Show threadthe roamer

@bluetea

Yeah, 100%. The poisoning of trust is the worst aspect of the whole thing. It voids any strategy that tries to maintain an LLM-free zone. It attacks us from inside the city walls.

#StopTheAICorruption

Mar 27 at 1:15amWeb
Show threadHelga18h ago
@the_roamer yeah it's a real issue. I'm dreading marking my next batch of essays - there's an obligation to maintain academic integrity, and I just don't know how I'm going to navigate it.
Show threadLooking for explanations…18h ago

@bluetea @the_roamer

And we know that some neurodivergent people write in a style that resembles AI…

Show threadHelga18h ago
@Susan60 @the_roamer this is where it gets problematic. A neurodivergent person close to me has been accused of using AI when I know for a fact that they didn't.
Show threadLooking for explanations…11h ago

@bluetea @the_roamer

Exactly.

Show threadsky17h ago

@bluetea @the_roamer One thing I've found when marking essays is that there's no reliable way to tell whether someone has used software to write.

BUT in the last couple of years students have started to hand in essays where the sources cited do exist, but they don't actually match the content.

So, for example, a student will write "Facebook has implemented new moderation techniques, including algorithm-based tone analysis (Agarwal, 2024, p. 12)"

but then when I go check Agarwal, this claim does not appear. So I check a couple of sources for each essay, and if they're not checking out do a more thorough check.

If it's a problem through the essay, I put it through our academic integrity system as "falsification of data/sources", rather than as "AI".

Perhaps in the process some students learn that "AI" isn't intelligent and doesn't produce reliable work.

Show threadHelga16h ago
@scroeser @the_roamer yeah I'm familiar with the mismatched sources issue, that's a big one. The subtle stylistic things and 'vagueness' are all a bit more difficult. It puts you in a difficult position - having to read 2500 words of AI garbage and marking it as if a human wrote it because you don't have enough evidence to be able to confidently call it out.
Show threadsky16h ago
@bluetea Frustrating for sure. Why bother going to university not to learn?!
Show threadscrewlisp18h ago

@the_roamer @bluetea

Did you see that now-famous rust article by the chief rust promoter at the rust foundation, that is this formulation over and over again:

I am going to invest in rust and make billions of dollars because rust is great
- quote, "every smart businessman"

I am going to retrain in rust because it is super the best decision for my career.
- quote, "every smart programmer"

I am going to have my product rewritten in rust because it is best
- quote, "every smart product owner"

Show threadscrewlisp18h ago
@the_roamer @bluetea
of which he clarified afterwards that he did not actually write the brain-meltingly stupid gruel, but the prompt used for generating "his" blog posts did stipulate to chatgpt "to write the articles in his personal style".
Show threadHelga18h ago
@screwlisp @the_roamer ha. Right there you go. God it's awful.
Show threadLisPi13h ago

@screwlisp @the_roamer @bluetea Meanwhile no stable ABI, static linking for anything interacting with the language directly, no specification, blessed language package manager cargo (let's encourage all the antipatterns), assumption of a single build tool/system (cargo), only a single implementation (that defines the reference documentation).

Anyway, "great" except for all the "features" that make it extra-vulnerable to ensloppification.

The modern packager’s security nightmare

One of the most important tasks of the distribution packager is to ensure that the software shipped to our users is free of security vulnerabilities. While finding and fixing the vulnerable code is…

Michał Górny