[suddenclarity]

On the contrary, I think they completely misjudged it. The AI initially worked as intended and that's why the product manager defended the flawed results on Twitter. What they didn't expect was that people would start generating Nazis to show how flawed it was. This caused backlash from people and organisations that they actually cared about.

I doubt the patched model will fix the underlying issue. They'll just apply exceptions to a few noteworthy historical keywords. They've been vocal about their filtering of image search for years and this is in line with that.