Pretty interesting to go from a world of deterministic code, to LLMs which can do incredible things, but unreliably. In a world of LLMs, I could imagine guardra

Pretty interesting to go from a world of deterministic code, to LLMs which can do incredible things, but unreliably. In a world of LLMs, I could imagine guardrails being a table-stakes part of engineering an ML system, just like unit tests, and CI/CD would be for traditional software.