Comment by Cyan488

Good spot, but the AI tells are getting so so much harder to spot.

If I had to guess, models have producing content that is, for most of the population, indistinguishable from reality for at least 6-12mo.

Very few people will watch a video frame by frame to pick up on or check for artifacts, especially in the places most of the population watch video (tiktok, Facebook).

I have no doubt that, just like the fast corrections to 8 fingers, the final edge cases like reflections are being ironed out in future model training.