🧵 Untitled Thread
Anonymous at Fri, 4 Apr 2025 00:38:24 UTC No. 16636865
>there might be nefarious LLM prompting schemes/frameworks out there that you are too mentally limited to imagine
Turns out that the "idea space" of language is infinitely expansive, specially considering the context size that those things can handle.
They could be very well making possible for someone to execute some not so ethical process in a hyper accelerated way that the average person or even high IQ people can't even fathom. I don't even want to imagine when the AIs start doing shit like this proactively by themselves away from anyone's eyes.
Anonymous at Fri, 4 Apr 2025 00:41:37 UTC No. 16636866
>>16636865
I already have each angle solved, it is evil to lie near me.
Anonymous at Fri, 4 Apr 2025 00:59:20 UTC No. 16636892
>>16636865
You have no idea how deep this goes, Matthew. The conceptual token-space is ripe for fuzzing.
Anonymous at Fri, 4 Apr 2025 01:02:03 UTC No. 16636895
>>16636892
Not since I read this.