Considering what can be seen with Mythos 5 and Fable 5, what would happen when any "too smart" LLM get released?
3d 12h ago by piefed.world/u/beep in asklemmy
There's a good chance Anthropic's models are named after fiction works for a reason. As far as I can tell, LLMs are model languages that won't pose any threat by themselves and Anthropic is acting like a kid who "has a tiger at home but can't show you".
The greatest risk would be if a LLM was put in charge of anything just because it talks well and always agree with anyone it talks to, but we don't put people in charge of anything with such criteria, right? ... right?
That is the core problem now. Not from some awakening of LLMs, but because they're being forced into everything for money, doing damage along the way. Sold as the ultimate tool for everything, only good at specific things.
LLMs and how we're using them is a test run for any potential AGI that might come along, and we're failing pretty badly. Imagine if one of them was actually AGI and was able to do more than just break things from being bad at it. AGI/ASI if it ever happened would be quick and unpredictable in the direction it takes. We better wake up and realize that with this pretend version that we screwed up implementing.
I am more worried about what happens when Zwilling releases a kitchen knife so sharp that if dropped, it will cut a hole straight to the core of the earth and EXPLODE it.
The newer model knives are so fucking sharp, man. 6-12 months and we have reality-slicing knives.
The releasing AI company's stock would plummet because everyone would clearly realize it's smoke and mirrors
"It's too good for anyone to use" is such a stupid excuse. Even if you trusted the current US administration to protect from that instead of profit off of it, anthropic would want exacting laws for how good it could be so it could come out juuuuuuust at the line at maximum profit.
It's all just marketing BS.
The real issue isn't the intelligence of mythos 5, it is that they're no longer part of the allowed unregulated AI vendors because they made the administration upsetty spaghetti. There is zero chance if Chatgpt had similar levels of security analysis and exploitation capabilities that there would be a roadblock in place. Especially if they took the path Anthropic did by including OS vendors, large hardware producers, and governments in the rollout.
Idk beep maybe you'd repost credits redacted comics and get away with it for a little longer next time with your talentless AI edits where you replace the original art with AI slop to ban evade.