I think the thing that makes me saddest about this whole “use an LLM to generate code” thing is that many of my heroes are no longer my heroes.
-
I think the thing that makes me saddest about this whole “use an LLM to generate code” thing is that many of my heroes are no longer my heroes.
@samir agreed
-
@sanityinc @lizzy Ron Jeffries is still one of my heroes, indeed.
@samir @sanityinc @lizzy —@RonJeffries is also one of my heroes!
-
@samir Proof once more that people don't have hidden depths, only hidden shallows.
@adrian I think there’s both! (I hope there’s both.)
-
-
-
I think the thing that makes me saddest about this whole “use an LLM to generate code” thing is that many of my heroes are no longer my heroes.
@samir I think I'm even more exhausted by the well-meaning, well-reasoned, balanced takes that try to find the positives in LLMs. I appreciate it and also usually try to argue in that way but specifically for this topic it's draining the life out of me
-
@samir I think I'm even more exhausted by the well-meaning, well-reasoned, balanced takes that try to find the positives in LLMs. I appreciate it and also usually try to argue in that way but specifically for this topic it's draining the life out of me
@dtemme Is it because most of these takes are in bad faith?
Because it’s definitely not in good faith when you ignore negative externalities (climate, etc.), power differential, the abuse required to make the product (model-training sweatshops)…
If anything, I think that one of the crimes here is that we technology people mostly ignored all this when it was in other fields, e.g. fast fashion.
-
@dtemme Is it because most of these takes are in bad faith?
Because it’s definitely not in good faith when you ignore negative externalities (climate, etc.), power differential, the abuse required to make the product (model-training sweatshops)…
If anything, I think that one of the crimes here is that we technology people mostly ignored all this when it was in other fields, e.g. fast fashion.
@dtemme Maybe that’s it. It’s not just that fast fashion, and fast programming, is bad. It’s that no one is arguing fast fashion is actually great and we should be doing way more of it. But for programming, there are so many people saying exactly this.
-
@dtemme Maybe that’s it. It’s not just that fast fashion, and fast programming, is bad. It’s that no one is arguing fast fashion is actually great and we should be doing way more of it. But for programming, there are so many people saying exactly this.
@samir agreed. Though I'm not sure I think of these as bad faith as such. And maybe that makes it feel worse for me.
I just can't get my head around why people are so eager to prove that their work can easily be replaced by the statistical average of the stolen works of the most prolific of their peers.
-
@samir agreed. Though I'm not sure I think of these as bad faith as such. And maybe that makes it feel worse for me.
I just can't get my head around why people are so eager to prove that their work can easily be replaced by the statistical average of the stolen works of the most prolific of their peers.
@dtemme I don’t get it either. Major despair.
-
@samir agreed. Though I'm not sure I think of these as bad faith as such. And maybe that makes it feel worse for me.
I just can't get my head around why people are so eager to prove that their work can easily be replaced by the statistical average of the stolen works of the most prolific of their peers.
@dtemme @samir When I speak to people after my talks, quite a few say „you’re absolutely right, we shouldn’t use this tech“. And then they talk about how they have been using and are going to be using this exact technology „reasonably“. Their values are so separated from their actions that they don’t even realise what just happened.
-
@dtemme @samir When I speak to people after my talks, quite a few say „you’re absolutely right, we shouldn’t use this tech“. And then they talk about how they have been using and are going to be using this exact technology „reasonably“. Their values are so separated from their actions that they don’t even realise what just happened.
-
@samir @sanityinc @lizzy —@RonJeffries is also one of my heroes!
@firepoet @samir @sanityinc @lizzy
Wow, thanks! I gotta say, though, you may be dipping pretty deep in the barrel of heroes to come up with me!
-
@samir Oh, yikes. That's disappointing. Even throws in "open source or open weights", like those words have any meaning in this context.
-
@dtemme I don’t get it either. Major despair.
@samir Despair here as well.
-
@samir my biggest objection to marriage, bigger than a feminist critique of it, is that people change
-
@samir the lesson I keep learning (am I?) is not to have heroes in the first place. But it’s hard and it sucks when they fall.
@samir @janl (this is strictly a "yes, and" to your point, to be clear) the way i've tried to think of it is that much like "good person", a hero is a particular moment in a person's life rather than some status that one attains. i have values, and various people over the years have taught me those or embodied them in particular moments, for which i admire them and am grateful. but later they might turn against those values, and i can criticize that without losing what's most important.
-
I mean, Deepmind’s Alphafold won the literal Nobel Prize in Chemistry, to universal agreement?
So Terence Tao and others exploring Deepmind’s ANNs for other sciences seems like a pretty obvious thing to do.
It’s not that all ANN’s became bad just because LLMs hallucinate.
And yes, the key difference is that in Science/Math you can have automated and/or experimental checking of results, unlike for code security, the statistically average Redditor, or MechaHitler
-
I mean, Deepmind’s Alphafold won the literal Nobel Prize in Chemistry, to universal agreement?
So Terence Tao and others exploring Deepmind’s ANNs for other sciences seems like a pretty obvious thing to do.
It’s not that all ANN’s became bad just because LLMs hallucinate.
And yes, the key difference is that in Science/Math you can have automated and/or experimental checking of results, unlike for code security, the statistically average Redditor, or MechaHitler
-
@samir
Simple. It's Anil Dash. He's nothing if not unreliable inconsistent.
