Do you think Tesla or Einstein with a LLM model, would spit out 5 new discoveries per week instead of one every decade?
Complaining about this is like complaining about the fact that people use IDEs to manage the imports in their Java class files. Ideally, you'd stop using a crap language that makes you wrap three lines of actual implementation in 50 lines of imports, declarations, and redundant-to-the-code comments to make doxygen (and a linter) happy, but if the institutional model prevents that... By all means use a good IDE to do half the work for you.
If his was just (ab)using chatgpt to create the papers from thin air then it'd be pretty easy for somebody to show where the pre-chatgpt papers are and the post-chatgpt papers are by the drop in quality. However, nobody has been citing the content of his papers when they make the claims about quality so I doubt chatgpt has an impact on the quality of his papers.
Who said the research was of flimsy quality?
Who said that the discoveries (as opposed to the BS padding journals require) was what LLM was used for?
They might or might not be. But not because he used an LLM to "polish them".
Something like physics, you can have a gap of months or years between a new theoretical breakthrough or the completion of a new machine giving you new raw data to churn through.
... but chemistry is something (no offense to chemists) you can do to the dirt outside your house, and every novel molecular arrangement can be paper-worthy.