It used to be kind of frustrating when its IDs were wrong, but lately almost all of the new mydas-fly observations are correctly IDed by users as Mydas clavatus using the auto-ID. It works!
This also means that for these observations, I can make them RG just by agreeing. As a result, our team of mydas-fly identifiers have more time to identify more observations without having to have two of us look at each observation.
Yes, despite the constant complaints about the CV, it keeps getting better. This positive experience of using a machine learning feature is a nice contrast to all the uproar about bringing in GenAI.
I can join in the praise! Traditionally the CV has a hard time with spiders, but in some cases it does a really good job.. of course the source material it is training on has a huge influence. But I have seen it getting better and more reliable once I/we went in and cleaned out observations from tons of false IDs (in all directions.. besides just going through the taxon and cleaning out wrong observations, it is also very helpful to look at the “often confused with” tab on the taxon page and have a look at those taxa as well).. for example we have a clean-up running for Tigrosas and other lycosid spiders of northern America and it already improved quite a bit I feel. The CV is not bad at all in getting the Tigrosa species right, despite them being pretty similar