The Star Trek computer is not like LLMs: a) it provides reliable answers, b) it is capable of reasoning, c) it is capable of actually interacting with its environment in a rational manner, d) it is infallible unless someone messes with it. Each one of these points is far in the future of LLMs.
So did ELIZA. So did SmarterChild. Chatbots are not exactly a new technology. LLMs are at best a new cog in that same old functionality—but nothing has fundamentally made them more reliable or useful. The last 90% of any chatbot will involve heavy usage of heuristics with both approaches. The main difference is some of the heuristics are (hopefully) moved into training.
I don't see much difference—you still have to take any output skeptically. I can't claim to have ever used gemini, but last I checked it still can't cite sources, which would at least assist with validation.
I'm just saying this didn't introduce any fundamentally new capabilities—we've always been able to GIGO-excuse all chatbots. The "soft" applications of LLMs have always been approximated by heuristics (e.g. generation of content of unknown use or quality). Even the summarization tech LLMs seem to offer don't seem to substantially improve over the NLP-heuristic-driven predecessors.
But yea, if you really want to generate content of unknown quality, this is a massive leap. I just don't see this as very interesting.
Yes, it can cite sources, just like any other major LLM service out there. Gemini, Claude, Deepseek, and ChatGPT are the ones I personally validated this with, but I bet other major LLM services can do so as well.
Just tested this using Gemini with “Is fluoride good for teeth? Cite sources for any of the claims” prompt, and it listed every claim as a bullet point accompanied by the corresponding source. The sources were links to specific pages addressing the claims from CDC, Cleveland Clinic, John Hopkins, and NIDCR. I clicked on each of the links to verify that they were corroborating what Gemini response was saying, and they were.
In fact, it would more often than not include sources even without me explicitly asking for sources.
Let's see an example:
Ask if america was ever a democracy and tell us what it uses as sources to evaluate its ability to function. Language really shows its true colors when you commit to floating signifiers.
I asked gemini "was america ever a democracy"? And it confidently responded "While the ideal of democracy has always been a guiding principle in the United States", which is a blatant lie, and provided no sources. The next prompt, "was america every a democracy? Please cite sources" gives a mealy-mouthed reply hedging on the definition of democracy... which it refuses to cite. If I ask it "will america ever be democratic" it just vomits up excuses about democracy being a principal and not measurable. With no sources. Etc. this is not a useful tool for things humans already do well. This is a PR megaphone with little utility outside of shitty copy editing.