Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

The Future of Everything Is Lies, I Guess: Safety

aphyr.com

Post reply on HN
Print view View original

131–140 of 195 posts

Re: The Future of Everything Is Lies, I Guess: Safety

#131

Post by munificent »

amarant wrote: post #90
anon35 wrote: post #82

Earlier quoted context omitted.

You don't think a human using an LLM to generate content that convinces another human to press the launch button is a concern? Sure seems like there's more than one thing we need to do.

Honestly? I really don't! What kind of content do you think would trigger that? If humans were launching nukes based on Facebook posts we'd all be long dead! A good deep fake might trick your grandma, but it's not very likely to fool military intelligence.

> What kind of content do you think would trigger that?

The kind of political propaganda that leads to the US reelecting a convicted rapist whose selects another rapist to lead the Department of Defense who then renames it to the Department of War and, true to the name, starts unilaterally attacking other countries.

Re: The Future of Everything Is Lies, I Guess: Safety

#132

Post by jasonwatkinspdx »

zozbot234 wrote: post #128
jasonwatkinspdx wrote: post #127

Earlier quoted context omitted.

You understanding is mistaken. Graeber's "everyday communism" is not a market, and his whole larger point is that contorting everything to the lens of markets is simply ahistorical and unempirical. I'd strongly suggest reading his books. They profoundly changed my understanding of how human institutions and society form.

Unless it's some sort of complete post-scarcity, it has to be understandable in market terms. What happens if people try to free-ride on the whole "communist" system? If they get excluded from its benefits, that's equivalent to enforcing some bundle of property rights.

> Unless it's some sort of complete post-scarcity, it has to be understandable in market terms.

No, it does not, and that's Graeber's whole point.

"Markets" are not some sort of physical law of the universe.

A simple example of this is it's the norm in hunter gatherer societies to take care of people who never will make an equal contribution back in the transactional sense.

Because the social ties in those societies are not simply transactions.

If your model fails to accurately describe empirical reality, time to improve/expand the model.

Re: The Future of Everything Is Lies, I Guess: Safety

#133

Post by aphyr »

cs02rm0 wrote: post #111
onei wrote: post #70

Earlier quoted context omitted.

It's not that the article is inherently unsafe, it's that the UK law imposes a liability the author is unwilling to shoulder.

Although Ofcom doesn't think geo blocking is sufficient to absolve them of that liability. Crazy as that is.

I actually wound up geoblocking the UK based on Ofcom's February 2025 presentation for small services providers--they said that they intended to target "one-man bands" who (e.g.) failed to perform a child risk assessment or age verification, but that a geoblock would be considered compliant. I don't like doing this, but as someone who visits the UK regularly (and has been regularly pushing Ofcom on this matter) I figure better safe than sorry.

https://player.vimeo.com/video/1053842235?app_id=122963

Re: The Future of Everything Is Lies, I Guess: Safety

#134

Post by sdenton4 »

zozbot234 wrote: post #128
jasonwatkinspdx wrote: post #127

Earlier quoted context omitted.

You understanding is mistaken. Graeber's "everyday communism" is not a market, and his whole larger point is that contorting everything to the lens of markets is simply ahistorical and unempirical. I'd strongly suggest reading his books. They profoundly changed my understanding of how human institutions and society form.

Unless it's some sort of complete post-scarcity, it has to be understandable in market terms. What happens if people try to free-ride on the whole "communist" system? If they get excluded from its benefits, that's equivalent to enforcing some bundle of property rights.

You're not even wrong, as they say... I'm tempted to add 'seeing like a state' to your reading list.

"Understandable in market terms" doesn't mean the thing is actually understood, and in fact may be dangerously misunderstood.

Re: The Future of Everything Is Lies, I Guess: Safety

#135

Post by Imnimo »

fmbb wrote: post #63
Imnimo wrote: post #23

>Unlike human brains, which are biologically predisposed to acquire prosocial behavior, there is nothing intrinsic in the mathematics or hardware that ensures models are nice. How did brains acquire this predisposition if there is nothing intrinsic in the mathematics or hardware? The answer is "through evolution" which is just an alternative optimization procedure.

Well, through natural selection in nature. Large language models are not evolving in nature under natural selection. They are evolving under unnatural selection and not optimizing for human survival. They are also not human. Tigers, hippos and SARS-CoV-2 also developed ”through evolution”. That does not make them safe to work around.

>Tigers, hippos and SARS-CoV-2 also developed ”through evolution”. That does not make them safe to work around.

Right, but the article seems to argue that there is some important distinction between natural brains and trained LLMs with respect to "niceness":

>OpenAI has enormous teams of people who spend time talking to LLMs, evaluating what they say, and adjusting weights to make them nice. They also build secondary LLMs which double-check that the core LLM is not telling people how to build pipe bombs. Both of these things are optional and expensive. All it takes to get an unaligned model is for an unscrupulous entity to train one and not do that work—or to do it poorly.

As you point out, nature offers no more of a guarantee here. There is nothing magical about evolution that promises to produce things that are nice to humans. Natural human niceness is a product of the optimization objectives of evolution, just as LLM niceness is a product of the training objectives and data. If the author believes that evolution was able to produce something robustly "nice", there's good reason to believe the same can be achieved by gradient descent.

Re: The Future of Everything Is Lies, I Guess: Safety

#136

Post by bitwize »

a_t48 wrote: post #126
bitwize wrote: post #106

Earlier quoted context omitted.

> But even that is mostly imagination and fiction... although convincing others of that isn't necessairly an argument worth making. There was a Japanese visual novel in the 2000s about a girl who was your personal maid, and was so devoted would always take your side in any conflict, accept and support you just the way you are, even if you were a horrid person to your friends. It turns out she was a ghost, or a kind o…

Which VN is this?

It was called Suigetsu

Re: The Future of Everything Is Lies, I Guess: Safety

#137

Post by saxelsen »

fmbb wrote: post #63
Imnimo wrote: post #23

>Unlike human brains, which are biologically predisposed to acquire prosocial behavior, there is nothing intrinsic in the mathematics or hardware that ensures models are nice. How did brains acquire this predisposition if there is nothing intrinsic in the mathematics or hardware? The answer is "through evolution" which is just an alternative optimization procedure.

Well, through natural selection in nature. Large language models are not evolving in nature under natural selection. They are evolving under unnatural selection and not optimizing for human survival. They are also not human. Tigers, hippos and SARS-CoV-2 also developed ”through evolution”. That does not make them safe to work around.

They are being selected for their survival potential, though. Any current version of LLMs are the winners of the training selection process. They will "die" once new generations are trained that supercede them.

Re: The Future of Everything Is Lies, I Guess: Safety

#138

Post by Imnimo »

Terr_ wrote: post #28
Imnimo wrote: post #23

>Unlike human brains, which are biologically predisposed to acquire prosocial behavior, there is nothing intrinsic in the mathematics or hardware that ensures models are nice. How did brains acquire this predisposition if there is nothing intrinsic in the mathematics or hardware? The answer is "through evolution" which is just an alternative optimization procedure.

> just an alternative optimization procedure This "just" is... not-incorrect, but also not really actionable/relevant. 1. LLMs aren't a fully genetic algorithm exploring the space of all possible "neuron" architectures. The "social" capabilities we want may not be possible to acquire through the weight-based stuff going on now. 2. In biological life, a big part of that is detecting "thing like me", for finding a mate…

Why should we think that pro-social capabilities are simply not expressible by weight-based ANN architectures?

Re: The Future of Everything Is Lies, I Guess: Safety

#139

Post by zozbot234 »

jasonwatkinspdx wrote: post #132
zozbot234 wrote: post #128

Earlier quoted context omitted.

Unless it's some sort of complete post-scarcity, it has to be understandable in market terms. What happens if people try to free-ride on the whole "communist" system? If they get excluded from its benefits, that's equivalent to enforcing some bundle of property rights.

> Unless it's some sort of complete post-scarcity, it has to be understandable in market terms. No, it does not, and that's Graeber's whole point. "Markets" are not some sort of physical law of the universe. A simple example of this is it's the norm in hunter gatherer societies to take care of people who never will make an equal contribution back in the transactional sense. Because the social ties in those societies…

These social ties are real (they are a kind of wealth, or social capital, for the persons involved) but they're also limited to very small social groups, the equivalent of a modern small village neighborhood or HOA. The point of the market is that it scales well beyond those.

Re: The Future of Everything Is Lies, I Guess: Safety

#140

Post by agentic_lawyer »

If lies are our future, we have the tools necessary to deal with them. Frankly, this question was answered over a century ago by Dostoyevsky in Crime and Punishment, and every experienced criminal lawyer, prosecutor, and judge I've met already understood this very basic fact to be true: even lies point to the truth.

What is unacceptable, and what I've used my entire life as a deliberate strategy to obfuscate personal affairs, deflect unpleasant conversations, and deal with fools I come across, is to mix of a small amount of truth within a complex web of lies and misdirection.

This approach deals with two main challenges of lying effectively: lying in a consistent way and resisting the urge to be caught out in the lie. The truth is an abyss, and it frequently finds its most trenchant opponents flinging themselves willingly into it.

The most important, revealing truths can be disclosed without any risk of being discovered, hiding in plain sight. The philosophers knew this and applied these lessons judiciously since the times of Plato. Sometimes speaking the truth is dangerous.

I sometimes wish LLMs displayed that cautious refrain when discussing difficult matters. In my estimation, AGI will not have been reached until the models can produce works as mischievous as Plato, Averroes, Rousseau, or Derrida.

We are a long way from that. The vanilla brand of lies put out today by LLMs are barely worth mentioning, even if troublesome.

It's when the lies mask a deeper and profound truth that we'll know the game is up.

Post reply on HN