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

OpenAI and Elon Musk

openai.com

Post reply on HN
Print view View original

351–360 of 1001 posts

Re: OpenAI and Elon Musk

#351

Post by andrepd »

TaylorAlexander wrote: post #275
elif wrote: post #56

Earlier quoted context omitted.

>they may be right that open sourcing AGI would lead to too much danger. I think this part is proving itself to be an understandable but false perspective. The hazard we are experiencing with LLM right now is not how freely accessible and powerfully truthy it's content is, but it is precisely the controls upon it which are trying to be injected by the large model operators which are generating mistrust and a poor und…

The biggest real problem I’m experiencing right now isn’t controls on the AI, it’s weird spam emails that bypass spam filters because they look real enough, but are just cold email marketing bullshit: https://x.com/tlalexander/status/1765122572067434857 These systems are making the internet a worse place to be.

Generative AI will make the world on general a worse place to be. They are not very good at writing truth, but they are very excellent at writing convincing bullshit. It's already difficult to distinguish generated text/image/video from human responses / real footage, its only gonna get more difficult to do so and cheaper to generate.

In other words, it's very likely generative AI will be very good at creating fake simulacra of reality, and very unlikely it will actually be good AGI. The worst possible outcome.

Re: OpenAI and Elon Musk

#352

Post by FeepingCreature »

BriggyDwiggs42 wrote: post #299
FeepingCreature wrote: post #258

Earlier quoted context omitted.

It sounds like you're just describing why our watch-and-see approach cannot handle a hard AGI/ASI takeoff. A system that first exhibits some questionable danger, then achieves complete victory a few days later, simply cannot be managed by an incremental approach. We pretty much have to pray that we get a few dangerous-but-not-too-dangerous "practice takeoffs" first, and if anything those will probably just make us th…

If there’s no advancements in alignment before takeoff, is there really any remote hope of doing anything? You’d need to legally halt ai progress everywhere in the world and carefully monitor large compute clusters or someone could still do it. Honestly I think we should put tons of money into the control problem, but otherwise just gamble it.

I mean, you have accurately summarized the exact thing that safety advocates want. :)

> legally halt ai progress everywhere in the world and carefully monitor large compute clusters

This is in fact the thing they're working on. That's the whole point of the flops-based training run reporting requirements.

Re: OpenAI and Elon Musk

#353

Post by Barrin92 »

Falimonda wrote: post #309
sharkjacobs wrote: post #297

> For example, Albania is using OpenAI’s tools to accelerate its EU accession by as much as 5.5 years What an insane thing to say so matter of factly. This is like a character in an airport bookstore political thriller who is poorly written to be smart.

Why is that such an insane thing for them to say?

because it's a completely made up claim with a random number thrown in. How do you end up with "5.5" years for a process that doesn't have a timetable, number of estimated emails required for EU accession divided by number of emails ChatGPT can generate per day?

Re: OpenAI and Elon Musk

#354

Post by FeepingCreature »

selestify wrote: post #317
FeepingCreature wrote: post #258

Earlier quoted context omitted.

It sounds like you're just describing why our watch-and-see approach cannot handle a hard AGI/ASI takeoff. A system that first exhibits some questionable danger, then achieves complete victory a few days later, simply cannot be managed by an incremental approach. We pretty much have to pray that we get a few dangerous-but-not-too-dangerous "practice takeoffs" first, and if anything those will probably just make us th…

What evidence do we have that a hard takeoff is likely?

What evidence do we have that it's impossible or even just very unlikely?

Re: OpenAI and Elon Musk

#355

Post by boredtofears »

BriggyDwiggs42 wrote: post #308
bergen wrote: post #277

Earlier quoted context omitted.

Just because something is not dangerous to the user doesn’t mean it can’t be dangerous for others when someone is wielding it maliciously

What kind of damage can you do with a current day llm? I’m guessing targeted scams or something? They aren’t even good hackers yet.

Fake revenge porn, nearly undetectable bot creation on social media with realistic profiles (I've already seen this on HN), generated artwork passed off as originals, chatbots that replace real-time human customer service but have none of the agency... I can keep going.

All of these are things that have already happened. These all were previously possible of course but now they are trivially scalable.

Re: OpenAI and Elon Musk

#357

Post by DrSiemer »

TaylorAlexander wrote: post #275
elif wrote: post #56

Earlier quoted context omitted.

>they may be right that open sourcing AGI would lead to too much danger. I think this part is proving itself to be an understandable but false perspective. The hazard we are experiencing with LLM right now is not how freely accessible and powerfully truthy it's content is, but it is precisely the controls upon it which are trying to be injected by the large model operators which are generating mistrust and a poor und…

The biggest real problem I’m experiencing right now isn’t controls on the AI, it’s weird spam emails that bypass spam filters because they look real enough, but are just cold email marketing bullshit: https://x.com/tlalexander/status/1765122572067434857 These systems are making the internet a worse place to be.

So you still believe the open internet has any chance of surviving what is coming? I admire your optimism.

Reliable information and communication will soon be opt-in only. The "open internet" will become an eclectic collection of random experiences, where any real human output will be a pleasant, rare surprise, that stirs the pot for a short blip before it is assimilated and buried under the flood.

Re: OpenAI and Elon Musk

#358

Post by jimkleiber »

> We are dedicated to the OpenAI mission and have pursued it every step of the way. (emphasis added)

> We intend to move to dismiss all of Elon’s claims. (emphasis added)

I feel somewhat surprised but more worried that people working in a space riddled with so much uncertainty seem to use so much certainty in their statements.

Re: OpenAI and Elon Musk

#359

Post by gitaarik »

advael wrote: post #113

Like most airing out of drama, this doesn't really materially change anything aside from making everyone involved look extremely bad But also, all entities involved looked bad already to observers who weren't so deeply rooting for them already that nothing would change their minds, so this is basically tabloid fodder drama in terms of importance as far as I can tell

Well, isn't it good that the public learns more about the bad stuff they're doing?
Post reply on HN