On 12 October, Engineering at Meta explained how it pools spare AI computing power across its fleet of servers for ads. When capacity is free, two things happen. Rank More Ads lets the system score more candidate ads before it picks one to show. Serve Better Models sends more requests to a more powerful prediction model. Meta says this means more relevant ads and better results for advertisers, but it published no numbers.
It sits alongside Meta's GEM post from 13 October, and both point the same way. Meta keeps handing more of the decision about which ad a person sees to its machine. The more candidates it can score, the more it matters what you give it to score.
My read on the agenda: better matching means more of every advertiser's budget performs well enough to keep spending, which is good for Meta's revenue and, if it works, good for you. The catch is that a smarter picker can only choose between the options in front of it. Five resizes of one ad give it one real option.
That's why this ties straight back to our Meta rule: creative diversity drives performance, and more angles beat more edits. You can't see this system or test it. You can control how many genuinely different reasons to buy sit in each ad set.