On the 8th, I queued up a handful of new photos for some wayspots. Just a few minutes ago, I realized I never actually uploaded them, so just did.
One spot had two photos, 45° degrees from two different sides (to capture the sides of the artbox).
One image went through AI and was approved immediately.
The other immediately went into In Voting (Niantic). I’d never noticed that happening before. Is that normal?
Just curious is all.
Approved:
In voting:
My standard experience for months now is to get a same minute accept on new photos. I am also seeing an uptick in submissions (photos and nominations) being sent to Niantic in house voting. My best guess (but just a guess) is that they are reviewing the decisions the ML model is making for quality control. I still tend to get those back accepted quickly.
Hmm, that could be what they’re doing. I’ve seen an uptick in my submissions going into Niantic Voting, and all tend to get approved within a few days. The majority of my submissions get approved by ML these days, too.
ML isn’t perfect, so it’s make sense to have some quality control measures in place to make sure it’s doing what Niantic wants it to be doing.
How Niantic’s ML works had just come up in conversation lately, and this article mentions reviewing the predictions:
@cyndiepooh is right. This is us checking on EMiLy decisions making sure the models make the appropriate decisions and teaching it back to her if she’s wrong.
Yep, it’s like an employee cooking up burgers at a fast food place taking the patties temps from time to time to make sure the griddle is cooking them to the proper temp so customers don’t get sick.
Can’t complain about quality control being done at all.
@NianticThibs if eMiLy approves something and then it gets reported for removal (and removed), or if it is denied than appealed and accepted, does that information get fed back to eMiLy to help it make better decisions in the future?
Really glad to know this is happening. Have been seeing reports here of people abusing the ML model to get stuff accepted that shouldn’t be. I would love to know what y’all are teaching it about neighborhood signs.
Only problem is the appeals team is completely unpredictable. They accept lots of things they shouldn’t and deny acceptable POIs for bizarre reasons. The ML should NOT learn from the appeals team.
