Second, most scientists aren’t academics (& the vast majority are not tenured academics). I actually think the most important product of academic research is not knowledge, but scientists. That’s why it needs to keep going, even under duress totalinternalreflectionblog.com/2023/05/30/w...

The takeaway of the paper isn't "Using ChatGPT to find and do exciting research makes your science suck." That's a separate question altogether, and not one this paper answers. n/n

This paper itself is a great example of excellent use of AI: a language model to find AI papers, another one to extract embeddings, another method to perform dimensionality reduction. The expanse of this paper wouldn't be possible without AI. 5/n

Their last result that AI adoption "is associated with contraction in knowledge." AI needs big data. and out-there research ideas and fields don't have big data, that's why they are out-there ideas. It's a tautological conclusion. 4/n

And yeah, AI papers take less time to publish than experimental ones, so naturally people who write AI papers write more papers and appear "more productive". Both of these things (doing research in a hot field and writing many papers) will correlate with career outcomes. 3/n

These methods pretty consistently outperform their predecessor statistical methods and are awesome.
The other findings are also pretty normal for any new field that gets hype (more citations). 2/n