What an asshole. He could have gotten the kid killed, not to mention the damage to his social reputation. And he can't even manage a "sorry if you were offended" non-apology.
Anyway, when he went after the Brown student saying he was "very likely" the shooter (also bringing in Mamdani again), he did less: he simply deleted the video.
Its failure mode are also vastly different. VLM-based extraction can misread entire sentences or miss entire paragraphs. Sonnet 3 had that issue. Computer vision models instead will make in-word typos.