We’re not at a point where this can be a human free pipeline, but I bet you could make this work with 1-2 humans in the pipeline that review instead of the 100 you would traditionally need.
These experiments fail because they greedily try to outsource all the work to AI. Another recent example: the lawyer who submitted ChapGPT hallucinations directly to a court case.
You don't need to eliminate humans, and certainly not at first. You just need to be much more efficient than status quo in order for AI to be deployed at scale.
> One article even failed to populate properly, with the text instead featuring a bracketed glimpse at how its opening sentence was supposed to read.
> "The Worthington Christian [[WINNING_TEAM_MASCOT]] defeated the Westerville North [[LOSING_TEAM_MASCOT]] 2-1 in an Ohio boys soccer game on Saturday," reads the butchered intro.
If the only data you have about a game is the sport, the team names, and the score, there's only so much an AI (or a human!) can do to write an interesting article about it. Once you've read a few dozen of the generated articles, they'll all start sounding the same -- because, aside from the details, they are all the same.
If you want quality articles, you need some more depth in the source data. And, for little-league sports games, that data may just not exist.
The numbers are probably meaningful to them due to their education though.
It’s like saying programmers spend their whole career pour though millions of lines of nigh random symbols they call source code.