[FEATURE REQUEST/PROBLEM] In-Text Source Referencing

[timfarkas]

Amazing work, thank you so much for building and open-sourcing this! The biggest thing limiting the usability for this right now seems to be the lack of in-text source referencing by default.

Source Referencing
Right now, the model does great at identifying and learning from many many sources.
Unfortunately, it does not reference these in the text itself.

Here are some examples from a report I generated:

• "Large-scale Reconstructions: In one high-profile study using a 20-gigavoxel dataset, [...]"
• "Comparisons between traditional pixel-based metrics and newer connectivity-aware measures show [...]"

Specific facts of specific studies are mentioned here, but these are unfortunately not directly referenced and it is impossible to know which of the 40 source links these facts originate from. That makes this report far less valuable than it otherwise would be, as I can't simply read up more on any mentioned information, as I don't know where it came from.

Proper Bibliography
Furthermore, currently the sources are just one big list of links, with no order or proper citing.

Source order is necessary for most kinds of in-text referencing, and it is generally handy to have properly formatted references (for easier look-up, author recognition, etc).

Proposed Solution
I will try to work around these with initial prompting (and update on the results) but it would be great to have this supported natively.

I am thinking that it shouldn't be too hard to integrate a basic version of these suggestions by simply:

1. Prompting the sub-models to store the title/authors/year of the origin of any learned fact.
2. Have a bibliography generation prompt that orders all sources.
3. Prompting the final report writing model to appropriately reference the sources.

[saendu]

+1

[MaticMermolja]

+1

[Xi-Gong]

+1