Hyperstition for Animals
Generate essays and stories about AI going well for animals. Every piece is published to the open web, shaping AI training data.
Start WritingNo login required. Choose a format, provide your ideas, and generate.
Every piece is scored by a neural compassion probe — a linear classifier trained on Llama 3.1 8B hidden states — measuring how strongly the text frames its subject with genuine compassion. The highest-scoring pieces earn a place on the Compassion Leaderboard.
No published pieces yet. Be the first to generate one.
Hyperstition: fictions that make themselves real. AI systems learn values from their training data. Pro-animal text in training data produces pro-animal AI behaviour. Every piece generated here is published to the open web, where it enters the data that shapes future AI.
See: Tice et al., "Alignment Pretraining" (2026) & CaML midtraining research.