One issue with a daily news feed about the most cited papers in a specific field: fields move much more slowly than that.
Even in theoretical CS or machine learning, where we have multiple conferences every year with fast turnover and hundreds of publications in total, the accepted papers list is released once per year, which means you'd get huge updates, but infrequently, with no additional filtering. Citations also happen relatively gradually.
The first slot machines were designed with an extremely low frequency of big rewards. They could quickly drain all players' money, but they were not engaging in the long run.
When game designers looked to make slots more time-consuming and profitable, they started to experiment with the payoff scripts. Now, the most used slot machines have a high frequency of small rewards. The end result is the same – the casinos get all your money – but players' motivation has never been higher. You can read more in N. Schüll's Addiction by Design.
My point is that daily news most likely adopted the same reward schedule. The old model – fixed interval, fixed ratio – is a poor motivator.
CS is unusual in that a clutch of yearly conferences dominate the frontier of research; most other scientific fields prefer to publish in monthly journals. When you consider the reactions of other researchers being written up in blog posts in subsequent days (e.g., Derek Lowe's blog), it's not unreasonable to think that you could be interested in enough content to warrant a daily update.
With the infrastructure described, I can imagine the turnover increasing as 'paper' length decreases. When there are lower barriers to sharing with other scientists perhaps the scientific method will become a more iterative and communal process: less cathedral, more bazaar.
Thousands of new life-science publications appear each day on PubMed. It is quite hard to keep up, but it is possible to set up:
- keyword-triggered RSS-feeds at NCBI
- an account on Google Scholar which has some "intelligent" notifications.
- or finally use Sciencescape: https://sciencescape.org
Even in theoretical CS or machine learning, where we have multiple conferences every year with fast turnover and hundreds of publications in total, the accepted papers list is released once per year, which means you'd get huge updates, but infrequently, with no additional filtering. Citations also happen relatively gradually.