11–20 of 24 posts
This is great. But the vast majority of non-gaming non-terminal-loving-developers human beings want to avoid the keyboard at all costs. They are doing it via trackpads, mousepads, touch screens, etc. Which are all inputs that transcend language or the ability to find meaningful words.
You could take it a step further and, instead of adding/removing buttons, simply updating the copy of (a) button(s) based on the question, which I imagine you could manage with a very small model.
Cool idea. FYI the GitHub link at the bottom leads to "page not found". Maybe the repo is not public?
LLMs generating just-in-time UI has a lot of interest and effort going into it. It's usually called "generative UI" or "dynamic UI generation". It was a pretty hot about a year ago. Here's a HF blog on it https://huggingface.co/blog/airabbitX/llm-chatbots-30 . Also check out Microsoft's Adaptive Cards. Nielsen Group wrote about it too. https://www.nngroup.com/articles/generative-ui/ The problem is it's hard to come u…
Main differentiator here is as-you-type UI customization, and also in this case the UI isn't generative in the sense the options are hard coded - the LLM just chooses between them.
I think we'll see a mix of all of the above in the future as we take full advantage of LLMs.
This looks like a neat idea, but I'm not too positive about it. This makes using computers even harder to explain to people who do not spend their entire day keeping up with the latest developments. They cannot form a mental image or reuse any memory of what will happen next, because it is all context dependent. On the other end of the spectrum, for power users, dynamically adapting user interfaces can also be quite…
My hunch is that typing 5 words could be faster than going through 3 menus and screens. I don't see this as a replacement for normal UI, but as an optional shortcut - if you're happy typing.
We're always coming back full circle: https://en.m.wikipedia.org/wiki/Ubiquity_(Firefox)
Thanks for taking your time to write this up
Neat idea. Regarding the performance, I think you could get a lot better performance by training a small classifier model, essentially an embedding model, and using the LLM as the distillation source. This would both be much smaller, addressing your desire for it to run in browser, while also being much more performant, addressing your quantization need. Using the full LLM is a bit overkill and you can extract the co…