You seem reasonable and even-tempered, so I’d like to give you a considered response so I can hear your thoughts. But first I wanna ask:

Do you object to the technology as a whole, or only to the use of the moniker AI to described LLMs?

What is your definition of intelligence?

I love the appropriate application of Kant!

But I think the point about bad faith misses the mark. For one to conduct themselves or argue for a position in bad faith (even against only themselves), they must first hold a true, hidden belief. I don’t think that’s typically the case here.

You could argue that there is a hidden belief and that it is something like “AI is bad” or “I don’t like AI.” However, the people arguing this position are often quick to point out themselves that AI has legitimate value, even if they then go on to narrowly define the character and application of that value.

I would argue instead that what is actually being hidden here isn’t a belief, but is instead a sentiment something like, “I’m scared.” Which, while being completely understandable, is an emotive state or pseudo-belief, not a truly-held but hidden belief. Because of that, they (in the general sense) aren’t calculating hypocrites; they’re are just panicking.

Although, I prefer the admittedly archaic flatus vocis for describing this particular brand of cognitive dissonance because it translates basically to “mouth farts,” and that makes me chuckle.

1. With greedy decoding (temperature set to 0) LLM output is deterministic. The same input will always result in the same output — every time. This is a fact.
2. Just because something isn’t amenable to calculation does not mean it is nondeterministic, for example the Three Body Problem. This is a fact.
3. Humans are stupid unpredictable base animals that hold almost no understanding of the world around them or their place within it, while at the same time their heard-level behaviors are almost, if not entirely, predictable. This is a cynical joke.

so there is a field of ai. … even partial steps have wide reaching applications and philosophical implications. it is the absolute coolest shit, and gets exactly 0 funding these days.

the thing all the data centers RAM water electricity and economy are backing right now is not ai. it’s what happens when a magic trick and a ponzi scheme fuck on a pile of cocaine in a Pharaoh’s tomb.

These two positions seem to be mutually exclusive, and your response is, on the whole, incoherent. Could you please explain what it is you are trying to communicate?

This is almost exactly the opposite of how LLMs work.

LLMs running under greedy decoding are deterministic finite state machines. A lot of work has gone into making their output nondeterministic because probabilistic next-token generation is more useful.

The more accurate characterization would be that LLMs are context-dependent deterministic state machines with adjustable nondeterministic sampling algorithms bolted onto the decoder.

Also, just as a side note:

They have no determinism, no concept of context and their “reaction” are pure statistics.

This almost perfectly describes human behavior.

They are statistical garbage generators with a high word prediction accuracy that is enough to fool humans into falsely believing they have understanding of context.

That very precisely describes a lot of people I know.

Could you inform me of what is really going on here?

I honestly don’t know what point you are trying to make, but since you are clearly confused by my comment, let me try to further explain.

My claim: The observation that you can’t have space exploration without ballistic missiles is correct. This analogy as applied to AI (specifically LLMs) is also correct, but falls short of perfect parity.

The gap — which is what I tried to address in my previous comment — is that for the aerospace example, space exploration vehicles are like ballistic missiles, but for LLMs those trained to do medical research or those trained to be a chat bot are often not just similar, but the exact same.

That was the point. An LLM built for research vs one built for chat are not just similar, like an ICBM vs a manned space vehicle, but are literally exactly the same object with no difference whatsoever.

This was my attempt to support the person I responded to against the critically uninformed, “Why don’t we just have the one (medical research AI) without the other (chat bot)?” crowd. With the obvious answer being, “Because they are the same thing.”

Q: So how’s it different? A: It’s not. That was my point.

Yeah, fair enough.

How is what people do any better or different than the Really Good Guessing Machine?

I really don’t understand this perspective, and it’s my biggest gripe with Chollet and his ilk as well. What, in your opinion, makes neuro-symbolic AI inherently superior to (the currently better performing) statistical methods?

Pretty sure that was “lost” as in, “needing a map.” Not everything needs to be about winning or losing.

You got a reference for that?

Yeah, that is a pretty shitty consequence, but it’s nothing new or different from the strandard clean room techniques the industry has been using for decades.

Does the cost outweigh the benefits?

No.

I fail to see how this has anything to do with the Harry Potter lady or the YouTube tech guy.

I think you are making a category error.

For your argument to be coherent, it must be the case that low-skilled individuals would otherwise go on to become domain experts if not for their use of large langue models. While the reality is that these individuals would likely never rise to the level of expertise in any case.

You and your interlocutor were essentially talking past one another. LLMs have a disproportionately negative effect on low-skilled, low-intelligence, or uncritical individuals. LLMs improve personal and professional development while being extremely useful to highly-skilled, high-intelligence, or critically thinking individuals. These two statements are not mutually exclusive.

The point that’s being danced around is that LLMs multiply the divide between those who blindly consume and those who think criticality for themselves, and this is a problem. The larger question of whether it shifts more of the later group into the former, is yet to be studied.

Yeah, I think you’ve nailed it.

There’s no stopping the Holocene extinction. Our only hope as a species is to try to survive it and to position ourselves so that we end up in the best-possible position on the other side. I see continued, rapid technological progress (including AI development) as the most realistic means to accomplishing these ends.

I don’t think Ishmael was wrong, I just think the viewpoint isn’t realistically actionable.

I think I’m (unfortunately) starting to really understand what Frank Herbert meant when he wrote, “Fear is the mind killer.” It’s very disheartening to see legitimate concerns snowball into lemming-like behavior.

When in the throws of despair or grieving for a society in decline I’d typically appeal to the classic MiB quote:

A person is smart. People are dumb, panicky dangerous animals and you know it.

But as the years go by it’s becoming increasingly difficult to find solace there — mostly because my experience is showing me that it’s simply no longer true. I feel like I’m living in a world where everyone read Bonhoeffer’s Theory of Stupidity and most people responded with, “Where do I sign?”

I’ve observed this behavior from both doomers and AI evangelists.

However, in the effort of full-disclosure, I must admit that I am a staunch accelerationist. That’s because, when viewed from a cosmological perspective, the only rational way to describe the last few hundred years of human development would be as an explosion, and I firmly believe the only path out is forward. Standing against the tide of technological progress at this point would be akin to standing against a blast wave.

That’s really damn scary.

The real sticking point seems to be that increased knowledge does not alleviate the existential dread. On the contrary, in many ways it exacerbates the mortal fear. Knowledge without power or prescience is impotent. Not that it would make a difference anyway.

If we had developed the concept of The Great Filter before the steam engine, would the world have responded to Watt’s machine in the same way it’s responding to Dario’s now? And would that have even made a lick of difference?

Except even moreso with language models. Here it is not just a similar technology, but the exact same vehicle.

Translation: The way scaling laws and gradient descent work for language models means that there is NO functional difference between a machine that does your homework and is designed to endlessly praise you, and a model specifically designed to identify viruses or medications.