This is borderline tyrannical and unreasonable. "Biases" can range from institutional discrimination to acceptable assumptions to literal facts. For example it isn't wrong to assume the strongest humans are men and the oldest people are women. But those are by definition "biases". Suppressing these are just outright denial of facts.
These "equality" pushes are getting dumb and dumber.
I don't think it is very high in the list of priorities, but I have no problem in making sure that those biases are not present in the ChatGPT output. This is not suppression of facts. The article is not talking about suppressing phrasing that right now you are more likely be a man if you have a Nobel prize, but it is talking about not assuming that you must be a man if you have a Nobel prize.
When you are trying to solve a crime, you only know that the murderer was a nurse, it's very important to assume a valid p(murderer|gender) as well as p(gender|nurse).
On the other hand, leaking p(gender|nurse) into a candidate scoring algorithm would be a no-no.
What people seem to ask for is very interesting actually. To both learn the underlying statistics AND learn not to use it explicitly in speech at the same time. Assuming next token prediction is the learned function, these two feel a bit contradictory.
(and obviously fixing the labelling of professor = man in text generation is actually much less useful for gender equality say than dealing with actual issues that lead to the P(X|category1)!=P(X|category2))