No confidential AI method is free. Each trades away speed, utility, hardware budget, or security. Compare five approaches for running LLMs on sensitive data: cryptography, TEEs, obfuscation, differential privacy, and hybrid splits.

Dive into the intricacies of Circle STARKs, their efficiency over Mersenne 31 prime, and the challenges of FFT/FRI compatibility.

Cryptographic commitments often bottleneck SNARKs. Circle STARKs and Binius reduce costs but add new challenges. GKR offers another path, cutting overhead by almost avoiding commitments entirely.

This post discusses different methods of constructing elliptic curves. It also covers some potential applications and practical considerations for developing curves for specific uses.

This cheat sheet provides a brief overview of the parameters that define an elliptic curve and offers a somewhat opinionated classification of existing elliptic curves.

Multi-party computation enables multiple parties to carry out a distributed computation of their secret inputs without revealing anything but the output.