JEPA And Omnilow

I'm currently reading a psychology book that discusses self-talk and language.

It made me realize that much of my thinking is not clean English. It is closer to compressed thought, where there is no exact English word or phrase for what I mean.

That led me to think about the JEPA direction associated with Yann LeCun's AMI Labs.

My current view is that JEPA-style world models and LLMs are not pure replacements for one another. They may connect through Omnilow-native LLMs.

In that kind of stack:

1. a JEPA-style model handles non-language representations
2. an Omnilow-native LLM structures and translates the reasoning based on those representations
3. until a premier LLM is Omnilow-native, a second LLM converts that output into more human-readable natural language

If that is directionally right, combining a JEPA-style model with an Omnilow-native LLM could allow natural language systems to handle very complex reasoning better than current LLMs can on their own.