That number, the 90%, is what I'm trying to base the discussion on.
Then you probably shouldn't be using the word "stereotypes." That term doesn't carry the connotation "correct 90% of the time."
Slight reframe: is it acceptable to assume something about members of a population, when that thing is true of 90% of the members of that population?
Let's invent an example and say that 90% of Lusitanians wear socks to bed. If you meet a Lusitanian, is it all right to ask them what type of socks they usually wear when they go to bed?
Alternately, a negative example. If the proportion of innocent people in jail is a randomly distributed 10%, and you meet someone who just got out of jail, is it reasonable to assume they committed a crime?
I would say yes in both examples, though depending on why you wanted the information, it would probably be wise to wait for further evidence. But knowing nothing else, you could start with a 90% probability and go from there.
Why are stereotypes problematic? Because the numbers are enormously skewed on the way to your brain from real life. Let's say that your "black people don't tip" example is actually true. What would that mean? Probably something along the lines of: 25% of black restaurant-goers are bad tippers, while 20% of restaurant-goers in general are bad tippers. Trouble is, your brain won't see it that way. Your brain, once the slight difference has been brought to its notice, will engage in confirmation bias and will notice cases where black people don't tip more than cases where black people do tip, or cases where other restaurant-goers don't tip. In other words, your brain will take a slight difference and elevate it to a rule of thumb. From "black people are slightly disproportionately bad tippers" to "black people are bad tippers."
Plus, your brain will go directly to conclusions, will not pass go, will not wait for further evidence.
This is why we are generally discouraged from stereotyping. It's true that stereotypes might have a seed of truth, and it would be false to claim that all stereotypes are wrong, or that there are no differences whatsoever between population groups, but our brains have so much trouble getting to the right level of nuance that it's probably better to avoid such things entirely.