The idea of doing this for research papers isn't bad, but I wonder how many people will pick it up. Depending on the subfield, I already see people act secretively about the current direction on which they are working, even with close colleagues. Sometimes it's about fear of being scooped, other times it's about trying a different direction. That would work against doing a stream of papers read, even if you could elect to share it only with some people.
Maybe more of an issue is the network effect. To really show value, I think they will need to capture enough people in specific sub sub areas that new ideas result, leading to publishable papers. Getting those people on to the service is a chicken and egg problem.