LLMs can be mind blowing, but thy have a fundamental flaw. they’re not grounded in real world observations. they’re trained on streams of text. World Models solve this. here’s how they do it: