Btw, he is talking about anomaly cancellation. One way to see it is when you use the lightcone gauge to analyse the EOM. After that you have to review the Lorentz algebra, because your treatment is not Lorentz-covariant explicitly. The Lorentz algebra is retained iff you impose certain conditions on the normal ordering constant and the spacetime dimension.

Note that choosing three spacetime dimensions also circumvents this anomaly.