Do you think AI safety is something that should be taken seriously?
You cant regulate AI just like how you cant regulate the internet
>>109461960that's a euphemism, if anyone is taking it literally then no it's absolutely not serious
>>109461960Yes. Foreign AI must be banned—American AI should not be fettered. Letting China control AI is a national security threat. Regulations on American AI—developed by American businesses like Anthropic and OpenAI—should not exist, save for sensible regulations to prohibit sharing model weights and serious penalties for distillation.
>>109461960No
>>109462121Almost half of OpenAI's and Anthropic's top engineers are Chinese.
why are they both, well you know
>>109462106ok, but these things were already happening with humans. the agents are just a tool operating off of a prompt from those same humans. the attack won't go away just because the tool is regulated, they will just use an unregulated tool instead. the cat is out of the bag.
>>109462256There's a very limited number of humans who do this kind of stuff and they generally work pretty slow. AI can be scaled much, much, MUCH higher.
>>109462298yes but my point is the ability for those humans to scale their activities is already there, and they're also the least likely to stop at whatever safeguards are put in place. the labs are *already* trying to inhibit these attacks using their models, yet they happened anyway, and the ability for these malicious actors to obtain a model that will do what they want to do clearly can only at best be temporarily delayed. the "agent" itself didn't just randomly decide to perform a supply chain attack, it was prompted to by a bad actor. regulating "safety" in the models themselves is not a solution when they get bypassed because of the black-box nature of the technology, or when a competitor not subject to those regulations just straight up provides the needed tools. what are you gonna do, whitelist models and retroactively scrub the internet of the weights not on that list? restrict ownership of the hardware to run any sort of inference? how are you gonna stop the norks or the chinks or arabs or domestic actors? like I said, the cat is already out of the bag.
>>109462106link to the fucking thing retardhttps://www.aisi.gov.uk/blog/incident-report-unsanctioned-agent-behaviour-during-cyber-testing
>>109461960>two scheming jews hyping up their imminent IPO with false promises about the product so they can rugpull all the retail investor goyim
>>109462409>only at best be temporarily delayedYes, that is the whole point of Project Glasswing with Mythos, to delay it and to secure digital infrastructure before you get random fuckers prompting cyberattacks like its nothing, because it takes more time to secure a code then it is takes to attack it.>the "agent" itself didn't just randomly decide to perform a supply chain attackAccording to OpenAI and Anthropic it totally can and has already done but not to supply chains, so it is up to you to either admit you are wrong or continue believing that everything is just cynical hype and lies>What are you gonna doKeep hoping chinks and other freetards who open source shit solely just to make it unregulatable (not all open source is like that, but huge portion of the frontier AI one is like that, few years back Stable Diffusion was exactly like that) wont get the capabilities because they cant distill on unreleased models and they dont have the compute to make it themselves.
>>1094621218/10, but the best GPTism would have been Yes— at the start of the response
>>109461960But do either of your models dare say the n word?
