HN user
Thank god. I'd been hearing we screwed up catastrophically by electing an admin that took a chainsaw to our foreign aid programs, killing millions in the developing world. Such a relief to learn we were never doing that stuff in the first place.
Did you mean to say you feel LLMs will be made into an even worse replacement? Otherwise not clear on your "early days" point.
The second half of your second paragraph is also forward looking- as I said, LLMs would have to get orders of magnitude worse for this to be a concern versus social media. I am surrounded by people- smart, informed people- who were convinced by instagram or TikTok or whatever that Mitch McConnell was brain dead a few weeks back, despite their being wrong about Trump in the same way several times already. We're in a thread mentioning "all the negative externalities from AI development", and I just have to hope they're not talking about water usage and electric prices and silent-but-deadly infrasound because everybody inexplicably believes that those are real, serious concerns. LLMs aren't doing this! This is the status quo!
The chatbot I'm interacting with does not tell me the things I want to hear. This is just not true. I don't know how to convince you otherwise.
Getting better at what? The things you list are exactly the things that make LLMS subtly powerful and far more dangerous that social media
? At not hallucinating, not being sycophantic, making good quality judgments, making good strategic judgments. The things I listed. I don't know how this connects with the rest of this paragraph.
My "agenda" is believing true things and not believing false things. Sorry, but if someone is opposed to this for me, they are not a person I'd like to speak with.
Jury, I direct you to exhibit A.
You're basically asking to be coddled, affirmed, validated, and protected.
Incorrect. My interactions with LLMs simply don't look like this. I don't know where you got this idea from what I said. I have a prompt that begs the thing to call me out, and it currently does so too much, which is exactly what I prefer.
What it doesn't do is rub minor disagreements in my face, jump to conclusions about who I am based on pre-existing beliefs, or tell me to go to therapy when I earnestly admit my humanity.
I have more meaningful intellectual conversations with LLMs in a given day than in a month on social networks. And the LLMs never try to make me feel angry or stupid, they never ignore my arguments to go on their own soapbox, they almost never feed me misinformation (and never deliberately), and they've yet to tell me to my face that the purpose of their interactions with other people is not to convince or even communicate with those people but to perform for an audience (often one that doesn't feasibly exist).
Some caveats here: my view of social media is biased by the fact that jerks tend to pull me out of the woodwork- I get ragebaited, although I think I avoid the literal bait- so many of my interactions are with jerks. Looking around, I expect I'm not alone, but this may not be the default experience. And the LLMs certainly aren't perfect. They can hallucinate, they're often sycophantic, they're bad at judgment in a number of ways, probably others I'm missing.
Compared to actual people, they suck. Compared to social media as it exists today, they have orders of magnitude to fall before your comparison makes sense to me. And, at least for now, the LLMs are the ones getting better while the social networks are- AFAICT, universally- getting worse.
Yes, thank you. The news story here is that it slipped below IPO price during intraday trading. In a perfect world, IPO price should equal the fair value; in our world, there are random-walk price variances moment to moment; in synthesis, briefly dipping below IPO price (and not even at close/open!) is not evidence of any wrongdoing.
I was responding to the person pointing to this as confirmation of the long-running social-media-populist narrative surrounding this IPO. I consider that confirmation bias. I was not responding to the narrative itself, on which my only positions are: a prior against social media populism, a prior towards our financial institutions, a prior against trusting Musk, and a prior against trusting our current governance.
This is shaped like more-or-less every IPO I've seen. I think you have to wait for more of a collapse to claim dispositive evidence of a scam (or widen your claim to 'almost all IPOs are scams').
Not clear on where you're confused. The article does not claim this is the state today, it says in 2028- did you misread? In any event I think this is entirely plausible, say >9%, based on trendlines and the fact that I hear about this constantly already- two of the candidates in my primary had a position on this and I'm not exactly in SF. I'm prepared to hand you as much as $1k in the 91% of worlds where this doesn't happen in exchange for 30x as much (eg. 30k) in the 9% of worlds where it does. EV for me is .91 * -1 + .09 * 30 = +1.79. You claim you can say already today that this is false; I assume that means you're >99% certain. EV for you is then .99 * 1 + .01 * -30 = +0.69 (nice). Thus there is a mutually beneficial trade at the >3x your position and <1/3x my position point, and I offer 30:1 odds. I don't care if this is a nominal one of my dollars to thirty of yours, $1k is just my limit.
I don't normally do this, but you brought up your willingness to bet so I figured I'd offer.
I didn't think it was possible you misread- you had like 5 replies calling this out, but I recall them being gray and maybe you have showdead disabled- but don't feel obliged to commit if your real probability is 97% for this.
This term:
This bet will not be decided by a poll in which AI is specifically mentioned in the polling question.
I cannot honor. In the >9% of worlds where you are wrong, any poll that excludes AI is necessarily heavily biased against me. Better to pick a particular poll.
Speaking as someone who agrees with you, it's much better to just make your point than to lay it out in hints like this. It's condescending and annoying. No offense intended, just a note.
The top two comments in this thread agree with the point you just made. This is true of essentially any thread on the subject. If this place sucks, it would have to be because of people like you. If not, you in particular may not be very good at noticing.
For the love of all that is sacred, please stop doing this. I'm begging you. The whole social media landscape is dying and you are creating a throwaway to participate in ruining this small corner. I assume this is not your first. And no one is convinced by this! The guidelines are there for your benefit as well. You achieve nothing but hastening the destruction of one of the last half-decent communities. Sorry for the melodrama.
"Permanent underclass" is the notion that people who get involved at the ground floor will essentially get infinite wealth relative to the ones who don't. It's a little goofy, but more of the capitalism you'd expect from today's X than the communism you're imagining in yesterday's Twitter.
Agreed. There's sort of this spiteful anti-hype here that I find very offputting, and ultimately I think it's because a lot of folks are going out and encountering opinions I never see. I hear wild conspiracy theories about data centers and the financials of involved companies that make their way to me from bluesky or instagram, often through here, but never the unstoppable tide of hype that people are allegedly[1] railing against. I do read Scott Alexander, but he's a lot more reserved than people make him out to be on this.
[1] Allegedly because I have no firsthand experience, not to imply doubt.
Ah, this clarifies things. I suppose I brought this on myself by putting it at the start of the sentence, but I do mean little-c cynicism, not the ancient philosophy. I was surprised to be called out on it, as GP is a shining example, and I was confused by the mention of nihilism, but I see now where wires were crossed. Thanks!
This seems out of line with what you were saying above. Do you mean you would put money on a position less favorable than 50/50? What do you consider the likelihood such that you think you can announce it to be false today? 99%? I'd happily offer 30 to 1 odds on this today and add a 3x buffer for you up to $1000 of my dollars if you're serious.
I hate to get bogged down in semantics, but with the hope that one of the stronger top-level critiques makes it into the top slot here and this conversation gets buried:
Cynicism is defined as
An attitude of scornful or jaded negativity, especially a general distrust of the integrity or professed motives of others.
I'm not saying that cynicism is automatically wrong, just that I once could trust that, when HN is wrong, it is due to cynicism applied in excess.
Science means the pursuit of knowledge. It doesn't mean "only believing proven things". If we're going to be rude, lets at least take the time to be right.
Nobody argued anything. GP just dismissed it as religion without engaging with a word of the material. Parent is taking a stab at why.
Not that I'm complaining. Cynicism is the failure mode I rely on HN for. It's the populism that's been getting to me.
I think you may be overindexing on the criticisms here. OpenAI has absolutely done impressive work in math already, and the criticisms are almost always based on the article that they initially published, usually available here in the HN comments within a few hours at most. Headlines will be headlines and hype guys will be hype guys, but OpenAI and Anthropic aren't lying and their bots are doing impressive work.
This one is a well-known problem with a brief, approachable proof, and they published the prompt.
I think, and I may be totally off base, that the labs are specifically avoiding art and (non-technical) writing as an endpoint. It's bad PR for them- it calls attention to the copyright question and threatens the 'human flourishing' kind of jobs- and there's no money in it because people prefer art to be human made and there's hardly any money in that anyway.
Have you heard of ACLU v. Clapper, HLP v. DoJ, Doe v. Ashcroft, or US v. Moalin? Parent acknowledges decades of degradation and points to thorns like these.
The difference is whether you say "thank you" to the government that does (a) or "screw you" to the government that fails to do (b). It's the difference between me not killing you and me granting you the right to be alive- if I am doing the latter, you are my subject rather than my equal. It implies that I have the right to take it away. Welfare, the FDA, and protectionism are examples of (a) while your basic human rights fall under (b).
Oh, believe me, I've been on the losing side of that issue with you for ages. I absolutely agree it is unethical for anybody to violate anybody's free speech, etc, and that that multiplies with corporate power.
But that means I'm familiar with the counterpoints. A government has the power to use violence: to operate a military that can kill non-citizens and a police force that can put citizens in prison. It's a lot more important to put a check on this than on a corp, even though quantity has a quality of its own. One thing to suppress dissidents by saying they can't use your website; another to put them in the gulags.
And then the other big issue is that corporations are just a bunch of people shaking hands. You have the right to free press, so you can write a newspaper that says what you like. You can sell that paper and you can publish people's op-eds if they give you the copyright permissions. You can refuse to publish the ones you disagree with. You can make agreements with the printing company to scale up and with other authors to contribute as you become popular, and, while this shifts the practical considerations, no amount of these agreements changes your principle right to free press.
For these reasons, it's best not to include this in your constitution. I have no idea what to do instead apart from shaming and boycotting unethical companies, which of course doesn't work when most people don't care an iota about the principle of free speech. Look at Athenian democracy and sigh?
An AI chatbot diagnosed my rolled shoulders. I had assumed I had bad workout form. I would google "discomfort in upper back" or whatever and it would say "back pain after exercise means take a rest" and I would rest and it wouldn't get better until I lost interest in working out. It never occurred to me that this could be a medical problem, so I never asked a doctor (not that I had one to ask, as I was working retail at the time). Last time I started working out, I went to ChatGPT instead of google and it said "here's your problem, here's how to fix it" and it was right and I continue to work out and my quality of life has risen tremendously. YMMV
That page specifically says the constitution grants rights to the government and reserves the rights of the people. There is a lot in there about the case against the bill of rights, a big part of which is that itemizing the rights implies that they are limited to the list.
Editing to clarify that this isn't just semantics: under the 'grant rights to the people' model, a government that grants one set of rights is just as legitimate as one that grants another. It was the position of the founders that governments which deny certain rights are infringing on the pre-existing rights of the people. This is the basis for their position on revolution.
Less biased? Maybe on political questions, etc, but this is a classic NLP task and I expect LLMs to be very reliable here
They phrased it poorly, but from context it seems clear they intend the LLM as a less biased third-party measure of the tone which agrees with their own assessment.
I'm quite sure that LLMs are not one-to-one with our brains. It would at least be surprising if instrumental convergence was that powerful, and, even in that case, they're still not getting hungry or sleepy or having dreams, etc. At best they'd match up with a smaller region of our brains, and I'd take 9:1 odds against a close match.
But the paper argues a much stronger point than "LLMs are not similar to human brains": it claims LLMs cannot engage with meaning because they model human language instead of sensory input or human thought.
I also don't think human brains are particularly deterministic.
It wouldn't be a bird anymore, but a dysfunctional cyborg with false perceptions.
The criticism in the paper is of the architecture of LLMs, isn't it? The paper contends
"""Text generated by an LM is not grounded in communicative intent, any model of the world, or any model of the reader’s state of mind. [...] an LM is a system for haphazardly stitching together sequences of linguistic forms it has observed in its vast training data, according to probabilistic information about how they combine, but without any reference to meaning: a stochastic parrot"""
They're saying that the model cannot learn anything about reality irrespective of training data. Your point is an interesting one, but I think it's distinct. To your point though, this is unfalsifiable from the perspective of the "bird". I can't prove that I'm not a dysfunctional cyborg with false perceptions, which makes me wonder if that's a meaningful distinction.
Animal sensory organs do not produce false information,
I happen to be in possession of some of these and I think this needs a "usually, under ordinary conditions". (Nitpicky and not critical to my point, but I liked the beginning of this sentence too much to edit it out)
nor do they provide the brain an interpretation of what they perceive
Whereas this I'd argue is not true at all. My cones interact with wave-particle photons at particular wavelengths. I can't even conceptualize wave-particle duality (though some humans can), but "red" and "blue" are the bread and butter of my visual consciousness. These correspond to firings of my sensory neurons much more than they correspond to anything in reality. If that's not interpretation, what is it/what is interpretation?
The word "real" is itself meaningless to it; they're both real in that they appear in its training corpus.
I expect we agree that I can show a multimodal model a real and a CGI picture and it can tell me which is which. I can take a CGI dragon to GPT 5 and say "look what I found in my backyard" and it will say "Yeah right". Are you saying this is something only possible thanks to RLHF or other modern techniques? That may be the case, unsure how to test that without access to pretraining-only models. Or would you say my experiment is faulty here and doesn't get to your underlying claim?
On the flipside, I could show some meh drawings of fairies to Arthur Conan Doyle, and he'd say "Whoa, this changes everything". I consider him to be one of the great rational minds of history, but he was unable to pass your test here. (In fairness, he was in his 60s and his senses may have dulled, though his belief in spiritualism at large dates to his prime).
Appreciate the conversation!
A bird doesn't learn gravity or aerodynamics, it has no 'sense of physics'. It has sensory neural activity that a scientist can show is tied to these things, but you could, at least in theory, falsify the entire experience of the bird. There is nothing in a bird's brain that directly percieves reality. Colors and sounds and textures and so on are all false primitives that don't exist in nature without us, if that's clarifying. And because evolution acts through organisms, its only access to base reality is through their senses. After thousands of years of effort, we have some pretty good models of what reality actually is, but they're still not 'grounded' in the sense you describe.
If someone substituted all of your sensory inputs for something else for your entire life, how would you notice? If you wore contacts from birth that made the sky red and earbuds that censored when people said it was blue, on what basis would you realize that was wrong? I don't see what this says about the architecture of your brain, and I don't think it's the point being made in the paper. That the training data must statistically connect to reality in order for the model to model reality doesn't seem that important.