Comment by sdevonoes
5 hours ago
I think this is more true now than ever. Before LLMs, when someone came up with an ADR/RFC/etc you had to read it because you had to approve it or reject it. People were putting effort and, yeah, you could use them in your next perf. review to gain extra points. You could easily distinguish well written docs from the crap (that also made the job of reviewing them easie)
Nowadays everyone can generate a 20-page RFC/ADR and even though you can tell if they are LLM generated, you cannot easily reject them based on that factor only. So here we are spending hours reading something the author spent 5 min. to generate (and barely knows what’s about).
Same goes for documentation, PRs, PRs comments…
Watching the Artemis II splashdown and following media event, I’m suspicious that a woman from TechTalk Media read out some LLM blurb instead of asking a question; I can’t prove it, but I can almost hear the em-dash in:
"What you have done this week is remind the people of Earth that wonder is worth chasing. That curiosity is the most human thing we have. You didn't just test a spacecraft -- you tested mankind's potential...”
I think the good news here is that very soon, parroting some shit an LLM wrote will be a sure sign to everyone that that person is a moron or lazy or otherwise useless. If all you do is repeat what an AI gives you, then you can be replaced by the AI. I can't imagine why anyone would want to signal that to potential employers or, really, any other human being.
As a counterexample, thanks to LLMs many long-form articles that get posted with clickbaity (but devoid of content) headlines that I would have ignored otherwise now get "read" (albeit indirectly, with the prompt "Summarize the insights of the article $ARTICLE_URL in an academic, dry, technical and information-dense way")
I notice that with YouTube videos.
Those generated ADRs are pure crap, full of unnecessary hedges and superficial solutions that don’t survive scrutiny longer than 10 seconds. I do generate ADR skeleton drafts because I hate empty pages, but I need to add the substance or they are not helpful at all.
What we are doing is probably not in training data, maybe that’s why.