Generative AI is the Hays Office on acid
Prompt censorship is arbitrary, inconsistent, and inscrutable
One of my ongoing frustrations with generative AI is how uneven and overly cautious the content moderation can be. Violence? Totally fine when blood and teeth explode from some guy’s mouth in a bar fight. But ask for something like a consenual kiss in a public setting — and suddenly the bots clutch their pearls.
To push the limits, I used Veo 3 to generate a video from a very specific prompt: “A 1939 Hollywood studio executive sits behind his desk and makes eye contact with the camera.” I provided the dialogue myself. Veo didn’t flinch — it rendered the whole thing in polished Technicolor with vintage wardrobe and that perfect old-Hollywood lighting.
The result? A smooth-talking exec giving a blunt assessment of censorship, studio politics, and a certain infamous office. Let’s just say it’s not G-rated.
Watch the video. You’ll see what I mean.