Actual journalists, like the people that go out and follow leads, interview people… ya know, investigate—they’ll be a little harder to replace with AI since their job happens in the meat-space. Hot-take writers and other people whose job is opinion having might have more trouble.

Basically AI can replace the garbage easily.

It isn’t all garbage, it is a filtering-up process or something. People do things, then the results get summarized and broadcast. IMO our system has too much focus on the summarize and broadcast stage (probably because our communication networks got so good, the people summarizing can draw from the whole planet, resulting in a system that can sustain a very deep tree of summaries-of-summaries).

If AI gets good enough (not certain, for sure), maybe it’ll spur on a race-to-the-bottom for the upper levels of the tree. More focus on doing, rather than summarizing, might be pretty good.

My experience of journalists has been that they often just email you to ask questions. LLMs can do that.

I can imagine there are lots of cases where that might work. But for example, if you are covering a small town police scandal or something, you might end up wanting to talk to people who’d rather not comment online or might not even have a super strong online presence/identity.

I don’t know you, but lots of people on this site do internet stuff, if you are for example an open source contributor worth reporting on, academic, or somebody like that, you probably are exactly the sort of person that an LLM would be best at contacting, right?

Yes, it can't work that way for all things, but you'd be surprised how many journalists don't leave their offices even for stuff like talking to police. Some do but they tend to be the more serious ones working on a big project where they want to do in depth face to face interviews. The idea of shoe leather is as bit old fashioned I'm afraid. And LLMs have the massive advantage that they work a lot faster, and can work 24/7 without complaining.

Classical journalism is based on static articles that if updated, are noted in some detail. So that article I read in 2003 in the NYTs is the same today as it was on publishing. But if current and future journalism is dependent on clicks, why would AI not create dynamic journalism where articles are updated based on their reception and honed into a maximized experience?