Redis Creator Debunks Claim That Chinese AI Models Just Distill US Models

Redis creator Salvatore Sanfilippo is pushing back against the narrative that Chinese AI models only get good by distilling American ones. His argument: you can't just copy a frontier model by hitting an API a few times.

Sanfilippo points out that a typical API only returns text answers. You don't get the probability distributions or internal states the model used to generate those answers. And without that, you're not replicating the core capabilities of a cutting-edge model.

The real reasoning and thinking power of a large language model is buried deep inside a staggeringly complex neural network. Through an API, all an external user sees is the final text output — not the full chain of thought or the probability calculations that produced it. It's like looking at a few test answers and claiming you've stolen the knowledge of the entire student.