Manifesto
The AI revolution was built on one obsession: predict the next word.Trillions of dollars were poured into that idea because it scales well on GPUs.
It worked better than anyone expected.Today’s models write, translate and explain almost anything.We think they have learned the surface of language, not what lies beneath it.More data made them more fluent.But fluency is not the same as understanding.
Humans do not think in words.
We think in concepts.We compress, abstract and reason.
Then we speak.
Today’s models do the opposite.They produce language first, and understanding, if it comes at all, is an accident.
We believe the order must be reversed.
To understand is to find the few ideas that explain the many things you see.A child who has met three dogs knows what a dog is.Intelligence is not how much you remember.
It is how little you need.
Understanding is compression.
We are building small models that think in concepts and understand before they act.We believe intelligence rests on a few simple principles.Our work is to find them.
We are teaching machines to think.