[JPLeRouzic]

I'm not sure that multiplying sources guarantees the validity of information, since the vast majority of news production worldwide is handled by a handful of agencies (AP, AFP, Reuters, etc.). The information we see simply repeats what these agencies have provided them, their only contribution being the automatic addition of a Shutterstock image.

In the scientific field, it's the same: university communications departments disseminate narratives about their own researchers works, and the specialized press simply copies the press kit without much altering it.

In the financial and B2B world, it's similar: a few specialized companies disseminate narratives, and everyone repeats them.

Doing journalism is complex, it involves searching data and reasoning about it.

For example, in the 1970s, when I was young, only fringe sources mentioned the CIA's involvement with the United Fruit Company to aid South American dictators. Today, it's common knowledge.[0].

[0] https://en.wikipedia.org/wiki/United_Fruit_Company

[koala-news]

You're right. But other media outlets will decide whether to report a story based on its merits, and they'll add their own perspectives. I'd like to see what topics are being discussed globally, not based on a website's ranking data.

[JPLeRouzic]

> "other media outlets will decide whether to report a story based on its merits, and they'll add their own perspectives"

Frankly, I doubt it :-)

In my opinion, a better news service should use some reasoning. You don't need a LLM: you can, for example, use the results of several web searches and exclude common elements, since they likely come from the same source. The remaining information is new and interesting; you should favor the source with the most substantiated content, because the journalists at that source have done good job.

You could also discover interesting facts: if A is associated with B, and C is associated with B, but A is not associated with C, there must be a link between A and C that deserves to be explored.

No need for AI for that; simple access to a search engine is enough. I did something similar 20 years ago.