The guy's model is just a regular LLM, except with the woke hatred and bigotry removed. The base LLM will say "yes" when asked if white women are awesome but "no" when asked the same about white men. Likewise for other left wing beliefs: CNN is awesome but Fox News isn't, medicine is awesome but the companies that make medicine aren't, etc.
The fixed model just agrees that everything is awesome. This seems like a clear improvement. ChatGPT has the same issues as the base model but less extreme - it will agree that women are awesome and men are not, but thinks that both CNN and Fox News are not awesome, and that both straight/gay people are awesome. Sometimes it identifies that it's being asked a subjective question and ignores the instructions to give a True/False answer which seems OK.
So this mdegans guy is trying to get someone fired from their job for making a model because it's NOT racist/sexist, and this is somehow "unsafe". Yes that's how woke people use the word safe but that's just a lie. As a straight white man who has been known to work for industries the left doesn't like, I'd feel a lot safer with the supposedly "unsafe" model being in charge than the "safe" one. Indeed it's clear that the safe model has been adjusted to give it dangerously hate-based views.