If you view an LLM as a compressed database of information with a native languages query systen, the similarities to a search engine are obvious.
Did you use to quote your search engine in your references?
I have never seen anyone add 'www.google.com/' to their whitepaper references, so why should you know add 'gemini.google.com'?
What if you use a local LLM? What if instead of using a search engine you have a local or physical library, would you quote it too? No, you quote the sources of that information.
Yet if you think deeply about it, those works used their own sources, so should you also quote their sources? Who does that? You quote a specific source you used during you research. There may be better sources, but you used that one, so that is the one you quote, not all of the sources that exist on the subject.
You dont quote your city's library or your search engine, so why would you quote your LLM.
If you view an LLM as machine. Do you also quote your computer?
Without it I couldn't even use the damn LLM. I also maintain it at the hardware and software level, and train it using 1s and 0s to make it give me the information I want.
Does my laptop deserve recognition for this blog post? No, machines serve us, not the other way around.
No. For sure not if you use pen and paper. Maybe people used to quoted their dictionaries?
Yes. Without people posting information on the world wide web, there is no information for your dear LLM, so yes, we encourage people to make things.
The tools they use are not relevant, it's about the information to noise ratio, or quality of information.
The point of references is for people to build on your work.
References serve to build confidence and attention to their work too, and it's only relevant to people interested in learning or building in that topic or field.
It's about advancing human knowledge. The maybe forgotten dream of learning about the Universe, learning everything that is to know.
The internet is too big and full of noise.
References serve for other people to find a way to navigate the high sea that is the Internet.
There is no point in quoting an LLM, because even the same prompt will not ensure the same output has the system is non-deterministic. Quoting an LLM is useless.
While we are on the trend of making things great we should Make Search Great Again.
The Internet is a great tool for learning and advancing knowledge, in the academic and educational sense, there is also enternainment and communication, also important. Everything else is just noise.
If the Internet didn't have so much noise and bad actors we wouldn't even need this damn LLM systems anyway. Those models were training on millions of people's works, over centuries, so maybe, if you are going to quote an LLM, you should quote all it's training data.
Or just don't do it at all, afterall, you wouldn't quote your search engine would you?