Yes! Important points. This is the key next step, in my opinion as well: giving it access to the exact text of primary sources. Something I'm working on and testing, in a variety of ways. ChatGPT4's newest model, o3, would seem to be a big step in that direction, but in my testing, it doesn't work as well as would have been hoped. Some of the deep research models work somewhat well in this regard. I'm excited to get access to Claude's new "deep research" tool, to see how it does in this regard
Yes! Important points. This is the key next step, in my opinion as well: giving it access to the exact text of primary sources. Something I'm working on and testing, in a variety of ways. ChatGPT4's newest model, o3, would seem to be a big step in that direction, but in my testing, it doesn't work as well as would have been hoped. Some of the deep research models work somewhat well in this regard. I'm excited to get access to Claude's new "deep research" tool, to see how it does in this regard
One project I’m working on is getting the major general-purpose LLMs to connect directly to Sefaria using their API.
There are a few technical challenges (CORS, mainly, so I’m working on setting up a serverless proxy function to work around them).
That said, things are moving so fast in this space that custom solutions like this might not even be needed much longer