Cassandra is only carbon nowNo, opposing LLMs isn't "purity culture." I've seen this now from quite a few different people, and I disagree vehemently. It is good, actually, to have moral principles and hold to them, even when people with more money than you find said principles annoying.
Jeff Atwood@xgranade it depends so much, I mean I can oppose screwdrivers being used to drive nails into the wall
@codinghorror Sure, but we're not talking about "which tool is best for driving a nail that I own into a wall that I own," we're talking about "is it ethical to use a technology built on fascist ideology and stolen work, that carries unconscionable environmental costs, and that's used to disrupt labor movements to perform a task that that technology is fundamentally unsuited to?"
It's quite fair to have a very firm "no" by way of answer to the second question.
@codinghorror Anyway, this isn't the first time you've replied to me to make the argument that LLMs are just another kind of tool. I suspect we won't see eye-to-eye on that, especially as my work has been abused to make LLM products.
I hope we can agree though, that my objection *even though you disagree with it* is principled and neither knee jerk nor purity culture.
@xgranade LLMs told me something critical about my health that no healthcare professional -- and I have a whole team working on me, because I'm bonkers -- ever did. If you want to ask, ask, I can provide very detailed citations and proof.
@codinghorror I'm not a doctor (well, not that *kind* of doctor, anyway), so I'll absolutely admit that I'm not the right person to evaluate those citations. I'll say that from a pretty damned nontrivial degree of expertise with machine learning, I would find it extremely surprising if *on average* text recombination without any underlying semantic model yielded useful advice more commonly than outright dangerous advice.
@codinghorror Like, nothing about LLMs and the theory behind them prevents anyone from getting lucky — and I'm glad that you got lucky instead of the much more common and probable case. But that doesn't mean that they're anything other than outright terrifyingly dangerous in a medical context more generally.
@xgranade people should absolutely be taught all the pros and cons, but I really dislike absolutism and zealotry.. it's not useful, it's not practical, it accomplishes nothing (except in the very narrow cases of civil rights or human dignity). If I wanted more ones and zeroes, I'd own more computers..
@xgranade and as I've said before, if you want to be angry, be angry at cryptocurrency which is gambling, grifters, and human trafficking to the bone. It's horrendous.
Chris Hanson@codinghorror @xgranade The push for LLM inevitability is all the same people as cryptocurrency. That should tell you something about LLMs. It certainly tells me something.
@codinghorror @xgranade See? You’re not a data point against the people pushing LLM inevitability, you’re taking a more measured approach than they are. ;) Those people are doing absolutely *insane* things like tracking LLM usage metrics and saying that’ll be factored into performance reviews. (Like is happening at Microsoft with Copilot.)
@codinghorror @eschaton Hey, don't put words in my mouth, I'm not part of that "y'all." I do not agree that doing propaganda work for some of the worst people on the planet, whether intentionally or not, counts as "measured."
But that's what you're doing right now by arguing in favor of LLMs.
@codinghorror @eschaton Given how messy this exchange has gotten, let me pull back slightly. I made a claim, that opposition to LLMs is not an example of "purity culture."
You, despite my explicit ask to not, came into my replies to make a separate but related claim: namely, that LLMs are sometimes useful, and implicitly that that utility is sufficiently great as to justify their ethical problems.
@codinghorror @eschaton While I explicitly said I didn't get into the second point, as the Discourse™ has gotten *incredibly* tedious by now, fine. You seem to insist on having that discussion out in my replies anyway.
To that end, I laid out several reasons that I find the claim that LLMs are "just a tool" odious: the euginicist origin, the fascist way they're funded and developed, that they attack and undermine labor, that they impose extreme environmental cost, and that they don't work.
@codinghorror @eschaton You've been very clear that you disagree with that latter point, and also that you expect I will find your disagreement compelling. I don't. It's an extraordinary claim that spicy autocomplete would produce the results ascribed to it, and that claim requires correspondingly extraordinary evidence. Anecdotes are a form of evidence, but without understanding the selection bias that goes into their collection, not on their own sufficient to show extraordinary claims.
@codinghorror @eschaton But fine, you disagree, I believe, as you've said earlier. I think you are very wrong on that, but I don't think either of us are budging on that right now.
Do you refute or disagree with the other points? Do you believe that there is some degree to which LLMs could, if they worked well enough, justify their usage given those problems?
@codinghorror @eschaton To be clear, I don't think you owe me any answers. I'm just one woman who's been doing this shit for decades, and who knows what the fuck she's talking about, but whatever.
It's that you made the claim *to me*, and have used that claim to justify that opposition to LLMs is pseudoreligous "zealotry." But you haven't addressed any of the substance of the opposition beyond putting forward one anecdote that I can't personally evaluate the veracity of.
@codinghorror @eschaton So far, the justification you've given for the "zealotry" comment has been almost entirely about the *shape* of the claims I made, almost without any reference to the *substance*.
This strikes me as a very strange way to approach other human beings and moral decisions in general.
Is there any strong claim that you would consider to not be "zealotry," or any degree to which a claim could be evidenced such that it would not be "zealotry" to you?
@codinghorror @eschaton Perhaps that's the root of our impasse, then. I fairly firmly believe that if something does that much harm to the environment, to labor movements, and to victims of fascism, it cannot be justified by appeals to its efficacy alone.
I suspect that if we cannot agree on basic moral precepts like "don't help fascists get rich" and "don't be a scab," there's probably not much hope for a favorable resolution.
Stephen 🌈 (he/him)@xgranade @codinghorror @eschaton After a few in-person conversations with folks who have gone “all-in” on LLMs, I think there is another effect at play. There appears to be a severe dopamine addiction happening. At least one friend of mine shows signs of it. And has said so. Likens it to social media addiction.
It is very hard to have rational arguments with addicts. So don’t worry if reason doesn’t always work. It isn’t your fault.
One friend has also mentioned how they are likely being managed out because of mild dyslexia. They can’t keep up with the sheer volume of text and code generated by the LLM and their colleagues who aren’t as neurodivergent can. It is very interesting to see this happening. They were a FANTASTIC programmer. And now are being put out to pasture very early.
It makes me very sad.
@firepoet @xgranade @eschaton it's the sycophancy problem more than anything else.. I've discussed at quite some length. It's really, REALLY dangerous and endemic at the "free" tiers particularly.. https://infosec.exchange/@codinghorror/116027959082917219