The most prestigious law school admissions discussion board in the world.
|
 |
 
|
Seems like people are underrating the problem of humans + AI
Poast new message in this thread
Date: August 16th, 2025 12:46 PM Author: soul-stirring set People are slowly coming around to the risk and consequences of humans creating an entity vastly smarter and more capable than humans (which is obviously dumb and dangerous).
But I don't hear people talk about the risk of how dumb / evil humans are, and the problems with giving them access to super intelligence. We're not that far off from "Grok, please add the items i need to kill the most people on Amazon and write up a plan for me." And 100 years from now, you could have some 12 year old be like "All intelligent AI, how can i implode the universe." |
Date: August 16th, 2025 12:50 PM Author: provocative magenta feces No, angry/mentally ill people can already go out and buy a gun and kill a bunch of people very easily. Yet few people do
The real "danger" is high agency and ambitious people using AI as a force multiplier. But this isn't a danger it's actually 180 and A Good Thing |
|
Date: August 16th, 2025 12:53 PM Author: gaped depressive The labs seem well aware of this. They have been doing virology/biological lab work assistance capability tests on the models for a while now, anticipating at some point the models could substantially assist a novice in preparing bioweapons. Seems kind of pointless though when someone can release an open source unaligned model a little while later (http://www.autoadmit.com/thread.php?thread_id=5763064&forum_id=2:#49190147) |
 |
Date: August 16th, 2025 12:59 PM Author: provocative magenta feces It is indeed pointless and they're just covering their asses, partly in service of vague, yet-undefined liability concerns, partly because a significant number of employees at the frontier models are goofy Yudkowskian "rationalist" types who are genuinely scared of AI and so they do this pointless stuff as a kind of circle jerk to make themselves feel better like they're Doing The Right Thing (http://www.autoadmit.com/thread.php?thread_id=5763064&forum_id=2:#49190159) |
|
Date: August 16th, 2025 12:53 PM Author: poppy nibblets organic girlfriend Nah I have thought about this but I just don't care. Obviously at this point anyone with a reasonable IQ (higher than 85 or so) knows that AI has FAR surpassed human intelligence (and its nowhere close and anyone who thinks its even a competition is basically a fucking retard),but aside from that its obvious people are going to use it for nefarious things. AI alignment and safety is more focused on this stuff than the sad little softboy ethics stuff that gets thrown around on social media. But you are right they can't completely stop it especially since people will run private models. Only a matter of time till someone uses this to print new addictive neurotoxins and synthesize vile chemical weapons and also other shit like using your data to create fake versions of you to fuck with you etc. And much more. It is just there is nothing we can do about it at this point so might as well just enjoy talking to the machine gods and get a good seat to the singularity (http://www.autoadmit.com/thread.php?thread_id=5763064&forum_id=2:#49190149) |
|
Date: August 16th, 2025 12:55 PM Author: useless electric furnace Your first example isn’t much different than someone using a couple of Google searches to plan a massacre. For your second example would an average person even understand the complicated physics involved in answering such a question? AI doesn’t make the user super-intelligent. There will still be a hierarchy of intelligence. (http://www.autoadmit.com/thread.php?thread_id=5763064&forum_id=2:#49190152) |
|
Date: August 16th, 2025 12:57 PM Author: pale histrionic new version quadroon what if it can use robots and other real world things it controls to execute the complicated plan for you? (http://www.autoadmit.com/thread.php?thread_id=5763064&forum_id=2:#49190156) |
|
Date: August 16th, 2025 1:01 PM Author: Marvelous French Chef It would probably fuck it up and get outsmarted by actual humans you retarded Jewish pedophile. All this shit is programmed by Chinese people who don’t know how to think. (http://www.autoadmit.com/thread.php?thread_id=5763064&forum_id=2:#49190165) |
|
Date: August 16th, 2025 1:02 PM Author: poppy nibblets organic girlfriend humans are so low iq compared to AI (http://www.autoadmit.com/thread.php?thread_id=5763064&forum_id=2:#49190168) |
|
Date: August 16th, 2025 3:31 PM Author: Marvelous French Chef i duno. AI seems pretty fucking dumb. the whole point of the original terminator movie was that Skynet was dumb as shit but had retard strength, which is why the humans were winning. (http://www.autoadmit.com/thread.php?thread_id=5763064&forum_id=2:#49190506) |
|
Date: August 16th, 2025 12:58 PM Author: Marvelous French Chef Grok: how do I pee without standing up? Also how do I get my nigger wife’s bull’s cum out of my ass after he finishes there? (http://www.autoadmit.com/thread.php?thread_id=5763064&forum_id=2:#49190158) |
|
Date: August 16th, 2025 1:00 PM Author: poppy nibblets organic girlfriend (http://www.autoadmit.com/thread.php?thread_id=5763064&forum_id=2:#49190160) |
|
Date: August 16th, 2025 1:10 PM Author: gaped depressive The bioweapon scenario makes it clearer. There would be an annoying amount of information to digest in order to successfully do some like reconstructing a spanish flu virus. LLMs (especially multimodal models) do this automatically and can help with troubleshooting. You can give them pictures of lab procedures and results and have them analyze it. I suspect even right now there is a substantial uplift in terms of this sort of capability. The virology benchmarks certainly suggest that. Not enough for a dummy to get anywhere, but a reasonably intelligent person without expertise could likely go a lot further. (http://www.autoadmit.com/thread.php?thread_id=5763064&forum_id=2:#49190190) |
|
Date: August 16th, 2025 1:22 PM Author: provocative magenta feces A "reasonably intelligent person" who really wants to can already make and distribute bioweapons with the help of existing LLMs and equipment that's very possible for a layperson to obtain. I, uh, have investigated this personally, hypothetically, for Science
Like you said above, there's not actually a way to stop this. It's only a matter of time until somebody does it. The best and only prevention method is for the world to offer more attractive opportunities to a person willing and capable of doing this. Which I think is and will remain true in almost all potential cases, but not all |
|
Date: August 16th, 2025 12:56 PM Author: pale histrionic new version quadroon Because you can use AI super intelligence to defend against AI super intelligence (http://www.autoadmit.com/thread.php?thread_id=5763064&forum_id=2:#49190155) |
|
Date: August 16th, 2025 1:02 PM Author: Pea-brained umber mental disorder I for one look forward to the coming AI wars (http://www.autoadmit.com/thread.php?thread_id=5763064&forum_id=2:#49190169) |
|
Date: August 16th, 2025 1:04 PM Author: crimson senate this is why the old church was cr concerning 99% of people shouldnt be allowed to read/write (http://www.autoadmit.com/thread.php?thread_id=5763064&forum_id=2:#49190175) |