EP187: Why is DeepSeek-OCR such a BIG DEAL?

Let me guess: companies like OpenAI, Claude, and Google won’t be keen on implementing this approach as they make their millions ($$) from token costs. Anyhow, converting text into compressed visual tokens seems like a clever way to significantly reduce number of tokens that reach LLM’s decoder layer. I hope DeepSeek’s OCR is incredibly accurate, unlike traditional OCR models.