I have finally seen the #iNaturalist species identification demo, #inat and #LLM peoples! And it's (surprise!) somewhere between just lame and genuinely harmful. If a student submitted this work, I'd give it a C+.

Given the hype, I was expecting better. I thought that maybe, if you train the LLM on some good data (which I would expect from most serious iNat users), it could at least generate a decent summary.

But nope. I did a spot check on a species that I know well, Christmas Fern, 1/X