Investigate use of LLMs to make safety documentation more accessible

[reiterative]

Goal

Evaluate the use of 'AI' tools trained using a constrained set of safety-related documentation as a way of making material easier to navigate, by supporting queries to summarise or find references to specific topics

[igor-stoppa]

I have tried https://notebooklm.google.com/ by feeding it the urls to the raw .md files from the Contributions directory and i was positively impressed. Unlike my experiment with ChatGPT, the Google LLM seems to really stick to the documents it was fed. Furthermore, whenever it generates an answer, it also provides references tothe text it is basing the answer on, so it is easy to confirm that it did not hallucinate.

Unfortunately, it seems impossible to share the custom chatbot I created, because it is tied to my gmail user.

However, we could provide instructions for people to create their own, by putting together a simple page with a link to the notebooklm and the urls of the documents to upload.

And that's it. It really is that easy.