Interesting and cool, but here's why serious academics should avoid it:

When you post Contributions, you grant us a license (including use of your name, trademarks, and logos): By posting any Contributions, you grant us an unrestricted, unlimited, irrevocable, perpetual, non-exclusive, transferable, royalty-free, fully-paid, worldwide right, and license to: use, copy, reproduce, distribute, sell, resell, publish, broadcast, retitle, store, publicly perform, publicly display, reformat, translate, excerpt (in whole or in part), and exploit your Contributions (including, without limitation, your image, name, and voice) for any purpose, commercial, advertising, or otherwise, to prepare derivative works of, or incorporate into other works, your Contributions, and to sublicense the licenses granted in this section. Our use and distribution may occur in any media formats and through any media channels.

Not person with original question but he asks for the same feature as I would like to see.

I am not sure how the documents are handled in your product, since Chat GPT has a context limit that probably wont be able to hold longer papers in memory.

For me, I have a pdf[0] depicting a system that can be programmed, along with bits of pseudo-code and a lot of clarifications. Something that the Chat GPT could use to spit out an actual implementation, if it were able to "think" about the pdf as a whole. I would love to see if your product is capable of such feat.

[0]: https://arrow.tudublin.ie/cgi/viewcontent.cgi?article=1177&c...

Drive-Based Utility-Maximizing Computer Game Non-Player Characters by Colm Sloan (note that basically only chapter 3 is needed in this case, but its still over 40 pages long)

Basic setup: point it at a folder of PDFs, have it recurse in and read them all, then ask it questions like:

* Summarize these papers on chimpanzee cooperation in the wild. What other papers should I be reading?

* Suggest an interesting master's thesis topic on the early modern economy.

* How good are polygenic scores at predicting educational attainment, and how has this developed over time?

Bonus: integrate it with e.g. google scholar, so it can go and find and read new papers.

Pricing: it is probably easier to start selling this to individual academics. Then when you've got a compelling product and the word is out, you could sell it to the whole lab (at a much higher price because people can put it in their grant budgets).

Gotchas: privacy. Nobody wants their hot unpublished paper to be scooped by a large language model.

Yes, this is the next step I'm looking forward to, and what would probably make LLMs really take off. Let me dump my own knowledge base or source code into ChatGPT and have it use that as its source of knowledge. I can only imagine the cost and resources required to train and run these individual models on a large scale must still be prohibitively large.