Proof systems / verifiable AI
Trace
Essays on proof systems and verifiable AI.
Visual essays on proof systems, transformer structure, and research in progress by Omar Espejel.
Latest essays
-
Why Transformers Fit STARKs
Transformer decode is a repeated state transition over carried context. That shape maps naturally to STARK-style traces.
-
Where Transformer Proof Pressure Appears
Dense arithmetic is not the whole zkML story. The interesting pressure appears in repeated softmax, normalization, range, and table-like work.
-
Proof Validity Is Not Statement Validity
For verifiable AI, a valid proof is not enough. The proof surface also has to bind the model, input, output, policy, verifier domain, and artifact.
-
Why Documentation Structure Fits Agent Workloads
A visual essay on how curated manifests, canonical metadata, crawler boundaries, and simple health checks make documentation easier for AI agents to retrieve and cite.