Tiered models separate public and private capabilities

Short summary

Tiered Language Models (TLM) use secret cryptographic keys to permute ~5% of parameters, enabling open-weight checkpoints to embed private capabilities invisible to users without the key. This framework eliminates the need for separate closed-API services, shifting IP protection from API monitoring to key management. Scaling to billions of parameters and key compromise scenarios remain unproven.

• •Secret keys route computation through private parameter branches in open-weight models
• •Single weight file can expose both public and private (licensed) configurations simultaneously
• •Avoids separate closed-API endpoints, moving security burden from service monitoring to cryptographic key protection

Generated with AI, which can make mistakes.