Eric Moore's Shortform
Oct 211
Safety can not be embedded only in model weights, it needs to be enforced by a cryptographically verifiable agentic harness. Most of the alignment field thinks AI models/weights can be reliably aligned to human values through training. I think this assumption breaks down as models become more capable. Any safety...
Hello! My name is Eric Moore, and since resigning from IBM (where I was the multi-agent consulting leader) in March 2025 I have been working on https://ciris.ai The app is on the app store and google play, and the alignment mechanism is at https://ciris.ai/ciris_accord.pdf Source code is at https://github.com/CIRISAI/ mostly...