There seems to be a notion that, when reviewing a paper, you are not supposed to use the knowledge acquired from that paper in your own research. In particular, you shouldn't build upon the paper's result, contact the authors to start a collaboration, etc. As spelled out in this answer, for instance: "once I finish reviewing a paper, I'm supposed to pretend that I don't know the paper exists".
I understand the justification for this when the work under review is not publicly available. But nowadays, in my community (theoretical CS), it is more and more common that authors will post their work as a publicly available preprint when submitting it to a conference. So the work that you review is publicly available -- in fact you may already know about it beforehand, or you can easily find this out.
In this case, the reviewer's "advance knowledge" of the paper is simply what anyone could get by reading the public preprint. Still, it seems that you are not supposed, for instance, to get in touch with the authors to start a collaboration with them on an improved result. So what is the remaining moral imperative for reviewers (if any) when reviewing papers that are publicly available? For instance, I would definitely cite the preprint if the work I'm doing happens to be connected to it. But I'm not sure, for instance, if it is a good idea to contact the authors to discuss possible improvements -- possibly without telling them explicitly that I know of their paper because I reviewed it.
(Note that this question is not about the importance of being impartial in your review, which is a different topic. For instance, it would still be an ethical violation to reject a paper because you are working on the same problem and want your own work to be accepted first.)