Sigh. OK.
-
@dnkboston @glyph @rmi LLMs are very addictive, by design.
The fawning behavior they default to is not mysteriously influential with people, and some people are more susceptible to that influence than others.
@RandomDamage I find it incredible that we haven't done anything meaningful to stop its rapid spread.
-
Like this is a luminary in the field, mocking his critics for making a prediction he finds risible when that prediction HAS ALREADY COME TRUE. He did the thing, and it went badly, in the way that everyone who is mad at him was saying all along that it would go badly. But he still thinks that he's right! This is not an unrepentant dumb guy being dumb, this is one of the smartest engineers in the field BECOMING dumb right before our eyes. That ought to scare you!
@glyph Many, many years ago, I used to celebrate each new blog post by Steve Yegge. What happened to him scared me. Now we have another victim and I'm more than just scared.
-
Like this is a luminary in the field, mocking his critics for making a prediction he finds risible when that prediction HAS ALREADY COME TRUE. He did the thing, and it went badly, in the way that everyone who is mad at him was saying all along that it would go badly. But he still thinks that he's right! This is not an unrepentant dumb guy being dumb, this is one of the smartest engineers in the field BECOMING dumb right before our eyes. That ought to scare you!
@glyph not enough people are talking about the early part of his post where he wrote (paraphrasing) that "no one knows" how human intelligence works, maybe we're all stochastic parrots. Motherfucker, read a book
-
@be oh fuck this _is_ channer shit isn't it. and it has that "raid" flavor of "unhinged chan trolls interacting with normies". Yeesh. I have registered my displeasure with certain LLM developments by participating in a few heated threads here and there but this is radioactively bad.
in any event thanks for the heads up
-
Sigh. OK. rsync discourse:
This is not a story about “vibe coding” or “slop” or regressions or even open source sustainability or whatever: it’s a story about mental health.
The timeline of Tridge’s response in particular can be broke down like so:
1. AI skeptics say "LLMs create difficult-to-evaluate defects, even if you're careful"
2. Tridge introduces defects even though he was careful
3. he gets yelled at
4. His response is to say "you dinosaurs don't appreciate how *careful* I was!"@glyph unpopular answer I am sure but this is a pile of biased bollocks.
This is a narrative built to prove an idea not a set of facts that leads to a conclusion.
Do better!
-
Like this is a luminary in the field, mocking his critics for making a prediction he finds risible when that prediction HAS ALREADY COME TRUE. He did the thing, and it went badly, in the way that everyone who is mad at him was saying all along that it would go badly. But he still thinks that he's right! This is not an unrepentant dumb guy being dumb, this is one of the smartest engineers in the field BECOMING dumb right before our eyes. That ought to scare you!
@glyph I actually praised his level-headedness cos of his blog but I didn't know he's THAT bad. I mean I don't really see smart people turn bad in real time. It's really rare for me.
-
@glyph I do understand the point being made and personally I'd much rather have software written without AI.
But. It's their project. Stop yelling at people for projects they make in their spare time. If you're mad, don't depend on FOSS, pay for your software.
No amount of good-intent behaviour, even if the outcome may not be great, in a FOSS software project justifies this sort of harassment.
@snaums @glyph That he maintains this software voluntarily and in his spare time isn't relevant here. I'm fully right to criticize his choice to destroy his reputation and his software, whatever it would be, FOSS or proprietary. If he doesn't want to be scrutinized for choices he makes for software he developes, then he shouldn't release this software in the public.
If you disagree, then would you still defend him with this rhetoric if he introduced malware in the code?
-
@glyph (as an aside - going to a casino with a bunch of math nerds, and watching the flashing lights make them entirely forget several years worth of statistics and probabilities is fucking fascinating. Convinced me to stop gambling for any significant amount of money.)
@miss_rodent @glyph The one sure way to win at gambling is to own the casino. The one way to capitalise on "AI" is to sell it.
Both are wrong to do in most contexts.
-
@KimSJ @crystallinefire @glyph @rmi
> a feeling of immense power and productivity.
Sure, and comfort, confirmation and reassurance alongside the confidence. And then for most of us, we start to spot the problems. It's not unlike working alongside a colleague who is super confident, but has little real world experience.
@digdilem @KimSJ @crystallinefire @glyph @rmi
> It's hard to get enough of something that almost works.
> - Vincent Felitti, MD (writing about alcohol addiction, but the analogy holds) -
@snaums @glyph That he maintains this software voluntarily and in his spare time isn't relevant here. I'm fully right to criticize his choice to destroy his reputation and his software, whatever it would be, FOSS or proprietary. If he doesn't want to be scrutinized for choices he makes for software he developes, then he shouldn't release this software in the public.
If you disagree, then would you still defend him with this rhetoric if he introduced malware in the code?
@raulmatias I’m not in the pro-LLM side of the debate, but the entitlement some people assume from OSS software isn’t less appalling.
Regardless of what the guy chose to do, read the license: he owes you nothing and yeah, you’re free to criticize, but if you don’t agree with him you’re also free to fork and maintain the code yourself.
Harassing an author for introducing AI slop into a verenable program is as moronic as introducing AI slop into a verenable program.
-
@glyph it always surprises me how a non-deterministic model could be taken seriously....
if you train it, due to its non-determinism it looses half its memory between hours of chatting with it. thats not a fault, thats *by design*
i can understand where hes coming from with "careful".
@biglinter @glyph they are deterministic, modulo precision issues which are VERY REAL AND UNDERREPRESENTED causing all kinds of chaos.
Production models generally run at a T≈0.7 which essentially does stochastic sampling to produce “creativity” or “variation” which can provide the illusion of depth e.g. if the model has a capability in its logits it may surface it given another run through as long as T is not 0.
I’m not saying this to argue, but because this misconception makes it even worse.
-
@biglinter @glyph they are deterministic, modulo precision issues which are VERY REAL AND UNDERREPRESENTED causing all kinds of chaos.
Production models generally run at a T≈0.7 which essentially does stochastic sampling to produce “creativity” or “variation” which can provide the illusion of depth e.g. if the model has a capability in its logits it may surface it given another run through as long as T is not 0.
I’m not saying this to argue, but because this misconception makes it even worse.
@biglinter @glyph what you’re referring to regarding context degradation is also a real problem, and important to understand, too.
But once you actually know about how these things work and solve problems it becomes somehow more insane that they work the way they do and the whole industry thought “this is fine, this is the technology’s final form”.
It’s so very half baked.
I look forward to a time when hopefully LLMs are a cute novelty for curious nerds in 10-15 years. Making your own is fun.
-
Like this is a luminary in the field, mocking his critics for making a prediction he finds risible when that prediction HAS ALREADY COME TRUE. He did the thing, and it went badly, in the way that everyone who is mad at him was saying all along that it would go badly. But he still thinks that he's right! This is not an unrepentant dumb guy being dumb, this is one of the smartest engineers in the field BECOMING dumb right before our eyes. That ought to scare you!
I hope to have more to say about this later, but this also demonstrates the weakness of considering intelligence as the only (or even primary) virtue.
-
Like this is a luminary in the field, mocking his critics for making a prediction he finds risible when that prediction HAS ALREADY COME TRUE. He did the thing, and it went badly, in the way that everyone who is mad at him was saying all along that it would go badly. But he still thinks that he's right! This is not an unrepentant dumb guy being dumb, this is one of the smartest engineers in the field BECOMING dumb right before our eyes. That ought to scare you!
@glyph It reminds me about how "good people" are falling for right-wing propaganda.
Many are highly intelligent, even world-class in their field; yet don't have an iota of social or emotional intelligence to tell that they're being manipulated.
-
@glyph It reminds me about how "good people" are falling for right-wing propaganda.
Many are highly intelligent, even world-class in their field; yet don't have an iota of social or emotional intelligence to tell that they're being manipulated.
@thomasfuchs @glyph Self-defined "pro-science skeptics" adopting and promoting anti-LGBT propaganda (especially the T) has been a problem for some years.
But as long as it didn't affect binary-conforming cishets, nobody was interested in investigating the phenomenon.
Moral: there are idiots everywhere and some of them are right bastards.
-
R relay@relay.mycrowd.ca shared this topic
