mesa-optimization

The quest for God's language—RL/GRPO on lossless information compression, super-prompting, xenolinguistics, semiodynamics, and context defragmentation in auto-regressive language models.

Minimal Mesa runtime for Intel GPU acceleration

Reverse-engineering an emergent mesa-optimizer: Transformer meta-RL on bandits, probed, patched, and stress-tested for the AI-safety failure modes it demonstrates.

Toy 5. An interactive proxy decay simulator showing how optimization pressure erodes the modeling capacity required to distinguish proxy from territory — producing self-reinforcing V(t) degradation that becomes progressively harder to correct. Companion simulation for The Depth Constraint — Series 2, Part 2.

⛔ [OBSOLETE] Q* implementation - Deprecated research direction

The Non-Separability Constraint: A unifying framework for understanding and detecting AI alignment failures