Antibodies are like the immune systems classifiers, a function that takes in a object and returns a probability that the object is a specific pathogen. If the antibody binds to the pathogen for a long time, then P_pathogen is high, if it quickly unbinds or does not bind at all, P_pathogen is low. Brownian motion is constantly bumping around all the tiny objects in the warm soup of your body, so each antibody function can be evaluated thousands of times per second.
In ML terms Omicron is like one of those adversarial image attacks that tricks a classifier into thinking a gun is a duck.
The interesting question is, was Omicron trained by nature or by man? https://bprice.substack.com/p/lab-leak-20