Where should user software[1] be going, by and large? E.g.:
You don’t mainly navigate a website by clicking around screens; you ask the LLM to bring up the relevant elements (currently hidden in arbitrary places several clicks away).
You don’t read large bodies of text such as wikis, you ask the LLM questions (and it shows you quotes etc.).
You don’t provide a UI at all, you provide data and tools for the LLM to use to do what the user wants (e.g. make a UI in their preferred style).
You don’t manage or ontologize data much, you just make it available for import and export across channels between LLM contexts.
You make infrastructure for caching / storing / quickly applied “crystallized” protocols / judgements made by slower LLM processes. (For example, the UI auto-reconfigures to make your normal workflow super easy.)
A cross-website/cross-platform note-taking system that allows you to make notes and comments—with the possibility of (selective) publicity or narrow sharing—so that you can compare notes and discuss things with other people. It may look like an optional moderated comment thread overlay that you can activate on any website.
I would also like to have something to do with highly customizable attention management. E.g., if I’m doing relatively high-urgency work, I would want to make going down relatively unrelated rabbit holes more effortful (e.g., the website starts glitching, or I have to type a lot of text manually to do this).
LLMs allow me to avoid software I dislike. I experience most web sites as a troublesome tedious UI sitting between myself and information I want to access. Now finally after about 25 years of increasing resentment towards web sites and web browsers, suddenly I can avoid most of the web sites I used to visit while accessing more online information than I used to.
(I took the trouble to install software that allows me to chat with LLMs without using a browser.)
Where should user software [1] be going, by and large? E.g.:
You don’t mainly navigate a website by clicking around screens; you ask the LLM to bring up the relevant elements (currently hidden in arbitrary places several clicks away).
You don’t read large bodies of text such as wikis, you ask the LLM questions (and it shows you quotes etc.).
You don’t provide a UI at all, you provide data and tools for the LLM to use to do what the user wants (e.g. make a UI in their preferred style).
You don’t manage or ontologize data much, you just make it available for import and export across channels between LLM contexts.
You make infrastructure for caching / storing / quickly applied “crystallized” protocols / judgements made by slower LLM processes. (For example, the UI auto-reconfigures to make your normal workflow super easy.)
What else?
Not sure what I mean; something like, software that involves a person interacting with something that’s not entirely local to their own machine.
A cross-website/cross-platform note-taking system that allows you to make notes and comments—with the possibility of (selective) publicity or narrow sharing—so that you can compare notes and discuss things with other people. It may look like an optional moderated comment thread overlay that you can activate on any website.
I would also like to have something to do with highly customizable attention management. E.g., if I’m doing relatively high-urgency work, I would want to make going down relatively unrelated rabbit holes more effortful (e.g., the website starts glitching, or I have to type a lot of text manually to do this).
LLMs allow me to avoid software I dislike. I experience most web sites as a troublesome tedious UI sitting between myself and information I want to access. Now finally after about 25 years of increasing resentment towards web sites and web browsers, suddenly I can avoid most of the web sites I used to visit while accessing more online information than I used to.
(I took the trouble to install software that allows me to chat with LLMs without using a browser.)