To people working in frontier AI organizations:

What if the values of the billions of humans directing AI cause unnecessary harm?

That is the gap almost nobody is working on.

We are building the largest documented record of good-values human behavior in real-world conditions.

This way, AGI learns from humans who treat each other well, not from the broken values that currently fill most of its training data.

This has to happen before AGI is built, not after.

Full argument here: https://provensuccess.ai/blog/human-values-problem-other-half-of-ai-safety

If you are working on this from inside a frontier AI organization, let's talk.