Pierre Bourdon

Hot take: if you have ethical or social problems with AI, you should not try to hide them behind quality problems. It's almost a certainty that quality will continue improving, and in fact even now a lot of the quality problems are avoided by a knowledgeable user of the tools. When that (imo inevitably) happens, your policies that were negotiated based on quality problems will be obsolete. If you entirely change your arguments at that point, it will be trivial to ignore you as unsatisfiable.

@ell1e@hachyderm.io as far as I know, all measurable metrics still show steady significant improvement for every new release. The obvious "they're gaming the benchmarks" argument falls apart given the amount of benchmarks close to real life usage tasks. And anecdotally, we see Fields medalists confirming that LLMs are close to having solved mathematical proofs. Vulnerability research is a field I know where it's clear 99% of performers are already outperformed. In many areas, quality is already solved.

@lina@vt.social that video linked in Wedson's retirement mail is... damning, to say the least. Wow. And the fact that this is apparently just acceptable behavior in that community...

@p4@masto.ai @danielleigh@mastodon.social @lina@vt.social which is especially funny coming from C programmers because... there's barely anything to learn, it's not a difficult language, and "experienced" C programmers aren't particularly good at avoiding the footguns that beginners also encounter.

@adrian@discuss.systems to be fair, that's kinda self-inflicted :-)

@tstudent@infosec.exchange why does your first paragraph not trivially apply to SW engineering if you define the unambiguous success condition as "passes review and QA"? Keeping some humans in the loop doesn't mean that LLMs can't wholesale replace human work in other tasks. Anecdotally, I have a bunch of ex FAANG colleagues whose SWE skills I deeply respect and who are now under LLM use mandate. The consensus I'm getting from them is that we're already past the "better than your average FAANG intern" point.