Debian Votes To Allow "Responsible Use Of Generative AI"
9 points by laktak
9 points by laktak
The operative word is responsible. What exactly is responsible? What happens when we accidentally oops? Just copied some copywriting code and we don't know it? With great responsibility. Comes great power.
It's a bit like "responsible use of heroin", as someone noted on Mastodon.
A powerful pain reliever in the hands of a professional, but also an unmitigated disaster when abused?
So it turns out doctors get booted out if they use the pain relief on themselves. It's for the patients! There isn't even any heroin in the supply cupboard!
Oh no, I agree, it's a wonky metaphor at best, but that was kind of my point. I am somewhat dissatisfied with the recent shift in AI-skeptic rhetoric, despite considering myself aligned a lot more towards them, than AI-bros and despite not wanting to use AI in my own off-work projects.
Recently, I've seen arguments shift from blaming the AI companies to ridiculing and denigrating users. Hell, I've seen completely serious parallels drawn between LLM users and nazis. My problem isn't raising concerns, there is plenty to raise and most of them require no value judgement of the person (resource usage, environmental concerns, worse code, etc.), but rather that the moment you say "you are in my eyes ontologically evil / corrupted", the well is completely poisoned and the other person will stop listening. They may not have listened in the first place, but there was at least the slim chance.
In my mind, the genie is not going back into the bottle and arguing that "no you can't use it or else you're bad" is not effective, because it offers no alternatives and actively turns the dutiful work of otherwise very well-meaning people into the very engine that makes them be taken less seriously.
A much better angle would be to 1) continuing to state the facts of just how much this all costs, because money is effective talk 2) working on methods that could mitigate impact in OSS (be that actually open models or at the very least strong rules how to make sure having an AI nearby doesn't bomb any project) 3) and, hopefully, bring AI providers into the court of law, so that even if they can't be forced to discard training data, its copyrightability or not could be at least established.
(And no, just because I focused on my grievances towards the AI-skeptic rhetoric, it doesn't mean the AI-bros aren't similar or worse. Silicon-valley dudebros, who want to get rich quick with AI, service-crippling scrapers and script kiddies, and those who can't put in minimal effort to at least check what they've generated while boasting that they made the next [insert famous and long established program here] are a lot more harmful, but these people also don't want to convince the anti crowd. They are at best indifferent, at worst actively hostile and so it's pointless to discuss how ineffective their rhetoric is.)