I think what he’s saying is that if the attacker has enough IP addresses at their disposal, they can spread out the attack broadly enough that any IP-based rate limit that would stop a bot would also impact human users. Thus most of the bot attempts slip in under the rate limit.