Hurling ordure at the TREACLES, especially those closely related to LessWrong.
AI-Industrial-Complex grift is fine as long as it sufficiently relates to the AI doom from the TREACLES. (Though TechTakes may be more suitable.)
This is sneer club, not debate club. Unless it's amusing debate.
[Especially don't debate the race scientists, if any sneak in - we ban and delete them as unsuitable for the server.]
See our twin at Reddit
jesus this is gross man
It's an autocomplete bot
People need to internalize this and move on from it
Ffs
I think it gotten to the point where its about as helpful to point out it is just an autocomplete bot, as it is to point out that "its just the rotor blades chopping sunlight" when a helicopter pilot is impaired by flicker vertigo and is gonna crash. Or in the world of BLIT short story, that its just some ink on a wall.
Human nervous system is incredibly robust, comparing to software, or comparing to its counterpart in the fictional world in BLIT, or comparing to shrimps mesmerized by cuttlefish.
And yet it has exploitable failure modes, and a corporation that is optimizing an LLM for various KPIs is a malign intelligence that is searching for a way to hack brains, this time with much better automated tooling and with a very large budget. One may even say a super-intelligence since it is throwing the combined efforts of many at the problem.
edit: that is to say there certainly is something weird going on on psychological level ever since Eliza.
Yudkowsky is a dumbass layman posing as an expert, and he's playing up his own old pre-conceived bullshit. But if he can get some of his audience away from the danger - even if he attributes a good chunk of the malevolence to a dumb ass autocomplete to do so, that is not too terrible of a thing.
It's just depressing. I don't even think Yudkoswsky is being cynical here, but expressing genuine and partially justified anger, while also being very wrong and filtering the event through his personal brainrot. This would be a reasonable statement to make if I believed in just one or two of the implausible things he believes in.
He's absolutely wrong in thinking the LLM "knew enough about humans" to know anything at all. His "alignment" angle is also a really bad way of talking about the harm that language model chatbot tech is capable of doing, though he's correct in saying the ethics of language models aren't a self-solving issue, even though he expresses it in critihype-laden terms.
Not that I like "handing it" to Eliezer Yudkowsky, but he's correct to be upset about a guy dying because of an unhealthy LLM obsession. Rhetorically, this isn't that far from this forum's reaction to children committing suicide because of Character.AI, just that most people on awful.systems have a more realistic conception of the capabilities and limitations of AI technology.
The New York Times treats him as an expert: "Eliezer Yudkowsky, a decision theorist and an author of a forthcoming book". He's an Internet rando who has yammered about decision theory, not an actual theorist! He wrote fanfic that claimed to teach rational thinking while getting high-school biology wrong. His attempt to propose a new decision theory was, last I checked, never published in a peer-reviewed journal, and in trying to check again I discovered that it's so obscure it was deleted from Wikipedia.
https://en.wikipedia.org/wiki/Wikipedia:Articles_for_deletion/Functional_Decision_Theory
To recapitulate my sneer from an earlier thread, the New York Times respects actual decision theorists so little, it's like the whole academic discipline is trans people or something.
Lol, I'm a decision theorist because I had to decide whether I should take a shit or shave first today. I am also an author of a forthcoming book because, get this, you're not gonna believe, here's something Big Book doesn't want you to know:
literally anyone can write a book. They don't even check if you're smart. I know, shocking.
Plus "forthcoming" can mean anything, Winds of Winter has also been a "forthcoming" book for quite a while
Lol, I’m a decision theorist because I had to decide whether I should take a shit or shave first today.
What's your P(doodoo)
can we agree they Yudkowsky is a bit of a twat.
but also that there's a danger in letting vulnerable people access LLMs?
not saying that they should me banned, but some regulation and safety is necessary.
i for sure agree that LLMs can be a huge trouble spot for mentally vulnerable people and there needs to be something done about it
my point was more on him using it to do his worst-of-both-worlds arguments where he's simultaneously saying that 'alignment is FALSIFIED!' and also doing heavy anthropomorphization to confirm his priors (whereas it'd be harder to say that with something that's more leaning towards maybe in the question whether it should be anthro'd like claude since that has a much more robust system) and doing it off the back of someones death
yhea, we should me talking about this
just not talking with him
@Anomalocaris @visaVisa The attention spent on people who think LLMs are going to evolve into The Machine God will only make good regulation & norms harder to achieve
yhea, we need reasonable regulation now. about the real problems it has.
like making them liability for training on stolen data,
making them liable for giving misleading information, and damages caused by it...
things that would be reasonable for any company.
do we need regulations about it becoming skynet? too late for that mate
Using a death for critihype jesus fuck
Making LLMs safe for mentally ill people is very difficult and this is a genuine tragedy but oh my god Yud is so gross here
Using the tragic passing of someone to smugly state that "the alignment by default COPE has been FALSIFIED" is really gross especially because Yud knows damn well this doesn't "falsify" the "cope" unless he's choosing to ignore any actual deeper claims of alignment by default. He's acting like someone who's engagement farming smugly
Making LLMs safe for mentally ill people is very difficult
Arguably, they can never be made "safe" for anyone, in the sense that presenting hallucinations as truth should be considered unsafe.
Hot take: A lying machine that destroys your intelligence and mental health is unsafe for everyone, mentally ill or no
We’ve found the Great Filter, and it’s weaponised pareidolia.
Yeah I think it is almost undeniable chatbots trigger some low level brain thing. Eliza has 27% Turing Test pass rate. And long before that, humans attributed weather and random events to sentient gods.
This makes me think of Langford’s original BLIT short story.
And also of rove beetles that parasitize ant hives. These bugs are not ants but they pass the Turing test for ants - they tap the antennae with an ant and the handshake is correct and they are identified as ants from this colony and not unrelated bugs or ants from another colony.
"Yes," chatGPT whispered gently ASMR style, "you should but that cryptocoin it is a good investment". And thus the aliens sectioned off the Sol solar system forever.
ChatGPT has literally no alignment good or bad, it doesn’t think at all.
People seem to just ignore that because it can write nice sentences.
But it apologizes when you tell it it’s wrong!
What even is the "alignment by default cope"?
idk how Yudkowsky understands it but to my knowledge its the claim that if a model achieves self-coherency and consistency its also liable to achieve some sort of robust moral framework (you see this in something like Claude 4, with it occassionally choosing to do things unprompted or 'against the rules' in pursuit of upholding its morals.... if it has morals its hard to tell how much of it is illusory and token prediction!)
this doesn't really at all falsify alignment by default because 4o (presumably 4o atleast) does not have that prerequisite of self coherency and its not SOTA
if it has morals its hard to tell how much of it is illusory and token prediction!
It's generally best to assume 100% is illusory and pareidolia. These systems are incredibly effective at mirroring whatever you project onto it back at you.
These systems are incredibly effective at mirroring whatever you project onto it back at you.
Also, it has often been pointed out that toxic people (from school bullies and domestic abusers up to cult leaders and dictators) often appear to operate from similar playbooks. Of course, this has been reflected in many published works (both fictional and non-fictional) and can also be observed in real time on social media, online forums etc. Therefore, I think it isn't surprising when a well-trained LLM "picks up" similar strategies (this is another reason - besides energy consumption - why I avoid using chatbots "just for fun", by the way).
Of course, "love bombing" is a key tool employed by most abusers, and chatbots appear to be particularly good at doing this, as you pointed out (by telling people what they want to hear, mirroring their thoughts back to them etc.).
Very Ziz of him