I want to be able to ask like 3, 4, 5 questions at once with these LLMs and have the response feel more fluid. Because at the moment it feels so high friction, they take too long, bumble around, I am impatient. And they respond with too much text. I don't want to read all that, I want it to be shorter, faster, grown, not refined. I don't want to dig, sift, sort. I want to ask and build up my understanding from ground truths, not filter out the slop to find the useful info.
Ngl some tool to paste a giant response and have it picked apart/sorted into something digestible is what I want. I have so much aversion to reading a wall of LLM words. Maybe it's just the friction of starting. Idk. How do I lower that?
See Reading.
Just "start" something
Maybe I really want the system to sort of just "start" something for me. Like premake a component, show me the essay skeleton. I want to see the options and choose the best (HAHAH THAT IS THE MCQ STUFF). It is just agency/effort + list of options/direction. Not planning psychosis hell.
See Induced fit and Growing an argument.
Smarter models
How many UX problems can just be solved with higher TPS and smarter models?
Eg I attach a file into context, why not just say "look" and find it, why do I have to bother knowing the file name?
Discovery
I think I want discovery and that's why I keep using Google search, it feels better to discover that way than an AI model. I think a big part of it is that I wanna see images lol. So any system we make should collate these!