Truth Curation: Combating Knowledge Drift in the Age of AI

Jason Kaufman | LavaCon 2025 – Atlanta

AI doesn’t hallucinate because it wants to. It hallucinates because the knowledge we feed it is often outdated, decontextualized, or incomplete.

At LavaCon 2025, I spoke about what happens when organizations move fast with AI but slow down when it comes to maintaining the truth behind it. The talk focused on a pattern we’ve seen over and over again: content that once served its purpose quietly drifts out of sync with the business, but still gets pulled into LLMs and copilots as if nothing has changed.

This is what we mean by knowledge drift. And left unchecked, it introduces risk every time AI is asked to support decisions, explain policy, or guide customer interactions.

In the session, I broke down three core ideas:

1. Truth is not static If the business moves, the knowledge has to move with it. Otherwise, AI starts serving answers based on assumptions that no longer hold.

2. Content needs context to be useful Information without source, scope, or verification history is noise. Especially at scale.

3. You don’t fix this with automation alone Human subject matter experts still matter. But they need workflows that make it easy to flag, fix, and approve changes in minutes, not months.

This is the foundation of truth curation. And it’s what we’re building for at Zaon.

You can watch the full session here: https://lavacon.adobeconnect.com/ptjw4gnrmq9k/

If your AI outputs depend on the quality of your knowledge, this talk might help reframe where the real work starts.