Comment by Rucadi | Hacker News Reader

Comment by Rucadi | Hacker News Reader

Parent Full thread

"Unsecured AI also has the potential to facilitate production of dangerous materials, such as biological and chemical weapons."

I think that we already have a good chunk of them, I doubt there could be worse than the ones we have.

"OpenAI worried because these text-generating AI systems can be used to generate massive amounts of well-written but misleading or toxic content."

This is solved by prestige, exactly what normative media had until some years ago, when they decided to throw everything away and just lick politics asses or just sell their integrity for money.

"You could ask them to design a more deadly coronavirus, provide instructions for making a bomb, make naked pictures of your favorite actor, or write a series of inflammatory text messages designed to make voters in swing states more angry about immigration. "

The coronavirus sentence is hilarious, and you don't need AI for any of that.

I'm kinda sick, what they mean with "regulating AI" is that they just want to control what the AI spits out to match their political agenda and beliefs, slowly impregnating the users of these AI of their political ideas.

Obviously, it's me and you who will not have access to uncensored models, big corps and the gov will have access to whatever they want unrestricted, and most hilarious, they WILL use the AI to "swing political beliefs".

If you need AI model to tell you how to make a bomb you are incapable of securing the materials and building one and now people want to “gatekeep” biological and chemical weapons as if the price of entry wasn’t already excessive to anyone who would have otherwise used these

> "Unsecured AI also has the potential to facilitate production of dangerous materials, such as biological and chemical weapons."

> "You could ask them to design a more deadly coronavirus, provide instructions for making a bomb [...]

Moreover, I call total bullshit.

I'd like for these people to show me how an LLM can instruct them to make a better virus.

Where's the hypothetical open source LLM that understands viral genomics? At the level of a PhD student?

Where?

Please show me. I'm waiting.

Isn't the point here the potential risk? By definition, nobody can point to a textbook instantiation of the risk because it hasn't happened yet, and the question is whether and when it might in the future.

It strikes me as similar logic to "you're saying I might get injured if I use this chainsaw without protective equipment? Well, show me the injury then!"

Outside of very specific workplace safety regulations, nobody is trying to regulate chainsaws.

Speaking of potentials, there's also a potential risk of outer space alien invasion, too. Yet our preparation here is appropriately scaled proportional to the risk.

Where is someone going to get a BSL4 lab with the necessary inputs to follow through with these wacky James Bond plots?

Maybe raise the alarm when splicing viral genes becomes a regular household occurrence. (And you don't even need to invoke the specter of LLMs for this scenario.)

Reply on news.ycombinator.com