Non-standard intelligence.

We build AI models on alien cognitive architectures. Reasoning systems that think in ways humans don't — and solve problems humans can't.

Models
BACKCHAT active

Reverse-causal language model. Speaks backwards — punchline before setup, conclusion before premise, end before beginning. Generates the destination first, then constructs the path that leads there.

KNOT active

Topological diffusion model. Generates through tangled latent spaces — ties, twists, and unravels structure in embedding space rather than denoising it. Outputs emerge from knot invariants, not Gaussian priors.

LIBER training

Large language model trained on occult, hermetic, and magickal texts. Grimoires, sigil systems, alchemical manuscripts, Thelema, chaos magick. Learns the grammar of the unspeakable.

Most AI labs optimise for the same benchmarks on the same architectures. We don't. humanity.tools explores the space of possible minds — building models whose cognitive patterns are fundamentally unlike human thought, then pointing them at problems where conventional intelligence fails.

We train esoteric architectures on strange data. The models are weird. They work.

so.long@thanks.fish