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

Sam and Greg's response to OpenAI Safety researcher claims

twitter.com

Post reply on HN
Print view View original

411–420 of 508 posts

Re: Sam and Greg's response to OpenAI Safety researcher claims

#411

Post by throwaway346434 »

iudqnolq wrote: post #406
mofeien wrote: post #396

Earlier quoted context omitted.

ChatGPT-4 won't kill us, and probably neither will the next generation of models. But if we keep building more and more powerful AI systems, eventually we will reach a point where one will become an existential threat. Present dangers of AI include Fake News, impersonation through deepfakes, biases and discrimination in the models, job loss and economic instability, loss of human purpose through individuals becoming…

If being able to write lots of convincing text could give you immense power why haven't any of the billions of humans that can already do that done so?

That's your misconception here - thinking it's one of billions. You have lived with spam and phishing emails your whole life. It's clearly not one bad actor alone, but by the same token it's not everyone in existence. Only what used to be one person running a scam is now one person running 10,000 automations to run a scam. Now make the automations more effective by 10%.

And that's the problem you don't even seem to recognise.

Re: Sam and Greg's response to OpenAI Safety researcher claims

#412

Post by ben_w »

bitnasty wrote: post #405
mofeien wrote: post #396

Earlier quoted context omitted.

ChatGPT-4 won't kill us, and probably neither will the next generation of models. But if we keep building more and more powerful AI systems, eventually we will reach a point where one will become an existential threat. Present dangers of AI include Fake News, impersonation through deepfakes, biases and discrimination in the models, job loss and economic instability, loss of human purpose through individuals becoming…

You, like many others, miss the number one danger of AI—making people dumber and less capable of reason.

People have been saying that about everything at least since Socratese was dismissive of writing; probably longer, but we can't tell because writing hadn't been invented yet.

Re: Sam and Greg's response to OpenAI Safety researcher claims

#413

Post by iudqnolq »

throwaway346434 wrote: post #411
iudqnolq wrote: post #406

Earlier quoted context omitted.

If being able to write lots of convincing text could give you immense power why haven't any of the billions of humans that can already do that done so?

That's your misconception here - thinking it's one of billions. You have lived with spam and phishing emails your whole life. It's clearly not one bad actor alone, but by the same token it's not everyone in existence. Only what used to be one person running a scam is now one person running 10,000 automations to run a scam. Now make the automations more effective by 10%. And that's the problem you don't even seem to r…

We started talking about the end of the world. Now we're talking about an increase in spam emails. I couldn't care less.

Spam calls in the USA skyrocketed in the past decade and society didn't end.

Re: Sam and Greg's response to OpenAI Safety researcher claims

#414

Post by JohnKemeny »

lucubratory wrote: post #313
JohnKemeny wrote: post #306

Earlier quoted context omitted.

There are (relatively simple) examples of what the transformer architecture is simply not able to do, regardless of training data, so that's simply not true.

Can you provide those examples?

On Limitations of the Transformer Architecture

https://arxiv.org/abs/2402.08164

Re: Sam and Greg's response to OpenAI Safety researcher claims

#415

Post by iudqnolq »

ben_w wrote: post #410
iudqnolq wrote: post #406

Earlier quoted context omitted.

If being able to write lots of convincing text could give you immense power why haven't any of the billions of humans that can already do that done so?

The most competent at that, did so, and are immensely powerful. That's basically all of politics, and not only in a democracy. You do have to be significantly above average to get there. Current generation LLMs are a bit above the human average, enough to be interesting, but probably not enough to be a personal Goebels for every would-be dictator.

Current LLMs are obviously far below the average adult human. Every demo I've ever seen brags about approaching the competency of an ordinary high school intern.

Even if they reach above average our whole society is structured around managing billions of above average humans. What's novel?

Re: Sam and Greg's response to OpenAI Safety researcher claims

#416

Post by _heimdall »

zer00eyz wrote: post #258
_heimdall wrote: post #169

Earlier quoted context omitted.

> None, and the court has already spoken on this. It's a pretty dead issue. The question of rights would be legislative rather than judicial. More importantly, the discussion should actually be among the people first in anything resembling a democracy, the few in charge are meant to represent us not rule us. > See PGE and its own death toll. Unless I'm mistaken, PGE isn't an artificial intelligence or sentient. > No,…

>> on the order of a couple years up to maybe 15 or 20 years Fusion, flying cars, AI in the 50's and in the 70's... Hell Hal was born in 97 and the film was in 68. For as impressive as LLM's are once you dig into them they aren't magical at all. A sophisticated model of language that predicts the next word is about as likely to become an AGI as whether predictions are likely to control the weather. Ask any expert in…

We don't have a clear, agreed upon definition of AGI, a way to test for it, or a way to predict whether a new model will meet those criteria.

The definition OpenAI uses for AGI is being more economically valuable than most humans at most things. That definition is entirely backwards looking. They'll keep releasing new models and occasionally check in to see how each model compared to humans economically. If that isn't bumble ducking our way into it I don't know what is.

Its interesting that you raise the difference in knowing the physics behind a bomb before building it and AI research stabbing in the dark as an acceptable thing. Its precisely that they don't know exactly what they're doing that there is so much risk. They're building potentially very powerful and dangerous things, connecting them to the public internet, and selling them to whoever wants to use them. That doesn't seem at all risky or dangerous to you?

Re: Sam and Greg's response to OpenAI Safety researcher claims

#417

Post by _heimdall »

francasso wrote: post #269
Silphendio wrote: post #182

Earlier quoted context omitted.

Child porn is already heavily "regulated". Producing the stuff is illegal and AI service providers must take steps to prevent it.

Are you saying they aren't already preventing it? I would find that hard to believe...

I'm sure they're preventing any prompts that explicitly ask for child porn, but they can't stop the inevitably difficult challenge of knowing when an image is porn of a child versus porn of someone that just looks under age versus graphic imagery of a child that may not meet some specific definition of what makes it porn.

Re: Sam and Greg's response to OpenAI Safety researcher claims

#418

Post by hamilyon2 »

I agree with many other commenters here. The danger of AI is not an uncontrollable superhuman.

The danger is much more close, down to earth and real.

Misidentified speech when we apply LLMs in chat moderation.

Wrong conclusions summarising 100-page military report.

More subtle things like hallucinating some details in news, which then after many years will be taken as historical fact.

The long tail of such consequences will be as big as LLM use. Maybe single issue will be much more impactful than the others, unseen ones. Death resulted from this will result in additional regulation. Other consequences will just become our new way of life

Re: Sam and Greg's response to OpenAI Safety researcher claims

#419

Post by d0gsg0w00f »

mofeien wrote: post #396
fullspectrumdev wrote: post #389

Earlier quoted context omitted.

> The technology they're working on poses the gravest of dangers Extraordinary claims require extraordinary evidence.

ChatGPT-4 won't kill us, and probably neither will the next generation of models. But if we keep building more and more powerful AI systems, eventually we will reach a point where one will become an existential threat. Present dangers of AI include Fake News, impersonation through deepfakes, biases and discrimination in the models, job loss and economic instability, loss of human purpose through individuals becoming…

I'm not that worried because humans are fickle creatures with an insanely sensitive trust/BS detector. One bad interaction out of 200 with a human/model and the human won't trust that human/model anymore. They'll categorize that human/model as a cute toy and look elsewhere for critical decisions or always get a second opinion.

My prediction is that digital content will end up just being "cute" and the real interactions/decisions will increasingly be offline and face-to-face. The open digital democracy that everyone hoped for will become tightly closed circles of humans talking face to face.

Re: Sam and Greg's response to OpenAI Safety researcher claims

#420

Post by RandomLensman »

ben_w wrote: post #410
iudqnolq wrote: post #406

Earlier quoted context omitted.

If being able to write lots of convincing text could give you immense power why haven't any of the billions of humans that can already do that done so?

The most competent at that, did so, and are immensely powerful. That's basically all of politics, and not only in a democracy. You do have to be significantly above average to get there. Current generation LLMs are a bit above the human average, enough to be interesting, but probably not enough to be a personal Goebels for every would-be dictator.

What are examples of immense powerful text writers these days?
Post reply on HN