There's an onus now on those that care about math to check the proofs. If errors are found in one their might be similar errors in another. A serial release would have been cleaner and allowed commentary that might help improve the accuracy or perhaps even the way the model conveyed the data to make it more understandable / digestable.

Openai dug the problems up from somewhere. They could have collaborated with some of the people that defined the problems in the first place. This is just perpetuation of their culture of theft.