Interactive citations with LLMs?

[DefenderOfBasic]

It would be way nicer if citations were more accessible as a UX:

This is sort of what citations are all about, but they could be a lot more interactive. Sometimes when I try to cite electronic sources with links for accessibility, traditional publishers will even make me take them out and cite a physical book instead.

https://x.com/christophcsmith/status/1815750583573250347

it sounds like there is a big need for this. There's maybe two sides of this? (1) interface/UX for the reader to peruse the book with its citations/referenced material (2) for the author to gather and collect references, linking them to relevant parts?

This is a really good idea for LLM use, actually. One of the things I've run into on the past when trying to make citations friendlier is that it's a lot of work. Many of my historian colleagues do a whole second book that's just a critical edition of their sources.

I like the UI of "who are we now" with the text on the left and diagrams on the right. https://whoarewenow.net/chapter-02/#ft-7. for sources, it's just static, but that should be dynamic. See Andy Matuschak's columnlar layout https://x.com/DefenderOfBasic/status/1814652781380178377

news sources

loosely related:

why don't news articles/opinion pieces publish their list of sources? Youtube video essays do that. I have never once read through it. but I think it's important that it's there

it can be messy, it can just be a little page with a link dump

imagine if one day a journal let you publish papers without citing anything. Chaos, how would we know anything was true? (why does TV news not include any way to look up sources) https://x.com/JeremyNguyenPhD/status/1803245454156763591

I think it should just be easier to "swipe to check the source"

https://x.com/DefenderOfBasic/status/1783264005395558838

[DefenderOfBasic]

for inspiration, see also "in a nutshell" by Nicky Case (trying to make embedded references in HTML an easy UX)

https://ncase.me/nutshell/

[chriscarrollsmith]

Wonderful, thank you! There are many dimensions to this problem:

1) Storage of sources (put them in a vector database to be exposed via a public API as part of the distribution model?)

2) Author's UX for reasoning about sources (RAG in the text editor?)

3) Author's UX for citing sources (automate finding and linking the relevant source material from the vector DB?)

4) Reader's UX for using citations (Nutshell-style expandable? Side-by-side panels? Open a new tab? RAG chat?)

5) Integration with already-published material (scan a page with your phone and search sources with an app?)

6) Distribution of new AI-native published material (probably has to be PC, tablet, phone, or AR/VR-based, but maybe you can do something mixed-media to supplement physical books with a digital device?)

Frankly, there are probably new licensing models required for closed-source texts you want to include in your vector db. This is a challenge but also a business opportunity to provide infra for this.