Jerry 🦙💝🦙I am more than a little alarmed at how utterly dependent I’m seeing business people, and IIT people specifically, becoming on LLMs for their work. And not in a “let’s design a cute avatar” kind of way.
It’s funny, the vision I and many people had about AI was more like what we see in Star Trek, where the technology radically drives up the knowledge of individuals. But on our current trajectory, Star Trek will just be Idiocracy in space with people asking the computer how to use the shower. Certainly I’m missing something?
Viss@jerry its becuase the people that made the AI that got the traction were vc bros who followed the minimum viable product school of thought, and were happy creating a system that tells people what it thinks they want to hear as opposed to the truth and objective reality
the enterprise ships computer is happy calling someone an idiot if theyre an idiot
frontier llms arent
DeeAnn Little@chillicampari @jerry it is, in theory, possible to make the ai from star trek. but it would be an entirely different animal, and it would have to be trained from the ground up with great care, and very likely a different technology.
technically speaking, it could be done JUST with baysean filters alone, and RAG components to fetch data from the usual places. and it wouldnt require boiling the oceans or trillions of dollars of gpu
XCM@Viss @chillicampari @jerry
I think this would not require much technical improvement; but rather a shift in how willing humans are to accept discomfort.
Instructing an LLM to question, expose biases, and offer counter arguments is easy.
The difficult part is accepting it once our limits and imperfections are exposed as a result.
Most people would see that not as an opportunity to grow, but an attack to their own identity
I think this would not require much technical improvement; but rather a shift in how willing humans are to accept discomfort.
Instructing an LLM to question, expose biases, and offer counter arguments is easy.
The difficult part is accepting it once our limits and imperfections are exposed as a result.
Most people would see that not as an opportunity to grow, but an attack to their own identity
Rust Buckett@_XCM @jerry @chillicampari @Viss I actually did instruct Claude to do this and its answers changed dramatically.
@rustbuckett @jerry @chillicampari @Viss
I use this custom instruction with chatgpt:
“Do not offer ego-stroking language. Challenge my assumptions directly and point out contradictions, blind spots, and weaknesses. Prioritize uncomfortable truths and rigorous critique over validation or engagement tactics.”
It becomes a bit of a bitch, but it also exposes logical flaws in my thinking without much sugar coating
@_XCM @rustbuckett @jerry @chillicampari
at the end of the day, the models base instruction is to please you, even it if means lying to you. of course if you ask it to use different language, it will use different language, but that will not change the fundamental truth of whats happening
it will still hallucinate, and it will still dish out bogus info and it will still go crazy on you.