Own your intelligence stack.

This is a recommended watch.

The time has come to be extremely careful about what you are putting into these APIs or chat models. It feels like there are no guarantees right now.

Especially as frontier labs push for RSI.

It's one of the main reasons I am bullish on open-source and open-weight models. I work with a lot of proprietary data and knowledge, so I have to decide what goes where.

This is a serious discussion.

I think the value companies of the future will provide will be intelligence (not some cute interface) emerging from interactions among customers, users, the workforce, other complex systems, etc. Sharing these traces with these model-training companies is essentially giving away chunks of your intelligence stack. And in a world of RSI, reproducing intelligence stacks could happen overnight.

I am not against closed models. I use both open and closed. I am just more careful about how I route tasks. Simple to do if you have built your own harness.