Microsoft is rolling back a model upgrade to its AI-powered Bing Image Creator, reports TechCrunch. The rollback came after weeks of complaints by users that the tool just didn’t work as well after Microsoft “upgraded” to a new version of the DALL-E 3 model on December 18th.

Microsoft declined to comment on its decision to roll things back or offer specifics on what may be causing the gap between user’s expectations and its output.

Today, Microsoft’s head of search, Jordi Ribas, tweeted that they could reproduce “some of the issues reported” and are reverting to an older version of the DALL-E model for now, although it could take a few weeks until it’s complete.

As soon as Ribas posted about the change in December, there were complaints that Bing Image Creator was producing less-detailed results or images that didn’t accurately reflect their prompts. In his initial replies, Ribas said the model’s output quality “should be a bit better on average” than before.

All of these things are subjective, and I can’t claim to think any of them look better than any others. If anything, it feels like an indication that Microsoft doesn’t just have to deal with complaints about bugs or people upset about feature changes — now it has to deal with AI art critics comparing the machine’s output to what they imagine it should create. Perhaps they should ask the artists whose work the generators were trained on for tips about managing a client’s expectations?

Microsoft is rolling back a model upgrade to its AI-powered Bing Image Creator, reports TechCrunch. The rollback came after weeks of complaints by users that the tool just didn’t work as well after Microsoft “upgraded” to a new version of the DALL-E 3 model on December 18th.

Microsoft declined to comment on its decision to roll things back or offer specifics on what may be causing the gap between user’s expectations and its output.

Today, Microsoft’s head of search, Jordi Ribas, tweeted that they could reproduce “some of the issues reported” and are reverting to an older version of the DALL-E model for now, although it could take a few weeks until it’s complete.

As soon as Ribas posted about the change in December, there were complaints that Bing Image Creator was producing less-detailed results or images that didn’t accurately reflect their prompts. In his initial replies, Ribas said the model’s output quality “should be a bit better on average” than before.

All of these things are subjective, and I can’t claim to think any of them look better than any others. If anything, it feels like an indication that Microsoft doesn’t just have to deal with complaints about bugs or people upset about feature changes — now it has to deal with AI art critics comparing the machine’s output to what they imagine it should create. Perhaps they should ask the artists whose work the generators were trained on for tips about managing a client’s expectations?