Please, just let me click

When you ask an AI for a book recommendation, it might respond with a follow-up: "What did you like most: the plot (A), the characters (B), or the setting (C)?"
In current interfaces, the user is then forced to manually type a response (like an animal), such as "Option A Please."
In any other system, making the user jump through hoops like that would be unwise.

Please, I don't want to type for that.
Natural language is actually a slow way of interacting, especially since much of the information found in a real conversation is lost in pure text form.

Instead, it would be incredibly helpful if the AI could generate or populate buttons for you to interact with.
In the book recommendation example, this would mean the user could simply click an option A / B / C.

This is very different from recommending follow up prompts and letting the users pick between a couple there. Those are very hard to predict and I think most of us don't even look at them since they're rarely helpful.

But simple buttons are just the beginning:
Imagine generating pictures and instead of you having to prompt for all of it, the AI also provides sliders for communicating possible changes, like "amount of detail" or "degree of ridiculousness." You could adapt the next iteration without ever having to type those instructions.
This should not be hard at all. You could just ask AI to do it (I did, you can try on the right).

This concept could be spun much further, too.
Models are so good now, you could easily imagine the AI making whole websites for each conversation, just so you can communicate your intent efficiently.
It's fun to think about taking this to extremes, but let's save it for another post.

One last thing: for simple tasks, answers should be preloaded for an amazingly fast experience, at least before Cerebras hopefully makes this all obsolete anyway.

A demo for this is coming soon!
I also already implemented parts of this in my RFP question answer demo, you can try it here.