The computer vision model on iNat is incredibly good. I mean, really amazing. And I can see the desire to have in words what the computer vision model is finding so that users can learn to ID too. But I think @inaturalist.bsky.social needs to figure out what it wants to be. - ThreadSky

About ThreadSky

markscherz.bsky.social • 10 days ago

The computer vision model on iNat is incredibly good. I mean, really amazing. And I can see the desire to have in words what the computer vision model is finding so that users can learn to ID too. But I think @inaturalist.bsky.social needs to figure out what it wants to be.

Comments

markscherz.bsky.social•10 days ago

If they want to be a field guide and have this information, there needs to be an infrastructure change that allows for keys, feature lists, trait-based ID, and taxon comparison. None of this really exists right now.

markscherz.bsky.social•10 days ago

This could be done with the help of genAI, but it will need expert validation, and ultimately will likely struggle to capture whatever it is that the computer vision model is actually using.

napoleonicento.bsky.social•10 days ago

I've been part of an experiment tonuse genAI to write an Extension short course. I've put in at least double the time fact checking amd reworking thr output than if I had just written it from scratch. So even if iNat had experts vet the output, ime it's still faster to not use genAI

markscherz.bsky.social•10 days ago

I also am an emphatic believer that iNat is systematically inflating identification confidence, because the current system is hard to implement fuzziness and unidentifiability in. And I am worried that genAI will make this worse, not better.

Posting Rules

Be respectful to others
No spam or self-promotion
Stay on topic
Follow Bluesky's terms of service

Comments

Posting Rules

Reply