In the same Colossus interview, Scott Wu brought up a question few people think about: what are humans actually better at than AI today? His answer isn't reasoning — on pure logic, AI has already won. What humans are really good at is retrieval — that ability to just remember something out of nowhere.
He used a phone verification code as an example. Two-factor codes are six digits; if you made them ten, most people wouldn't be able to remember them. The human brain's "context window" is just that short. But there's one thing the human brain can do that AI can't: you open a code file, and suddenly your mind surfaces a related bug you fixed four months ago — what happened, how you fixed it, why you made that change — all at once. Wu calls this "soft context": not verbatim memory, but the experience you've accumulated over years of work, precisely triggered by a cue.
He says that in his daily decisions at Cognition, he doesn't derive things from scratch like a logic problem. Instead, his brain automatically pulls up implicit judgments built up over time: "Why did we design it this way?" or "After that new technology came out, we decided to migrate to another framework." AI's context window is vastly larger than a human's, but when it comes to associative retrieval across time and scenarios, it's far behind.
AI is chasing this capability along two paths. One is embedding search — pulling relevant snippets from massive historical data, essentially a direct imitation of the brain's associative memory. The other, more radical route is continual learning, where AI updates its own weights while working — similar to the neuroscience principle that "neurons that fire together wire together." Wu thinks the second path is closer to how the human brain actually works, but it's still very early.
Near the end of the interview, he said: "At the end of the day, doing AI is really about solving cognition itself. Humans have spent thousands of years speeding up everything, but one thing has never changed: you still need a person to think and invent. Now even that premise might be rewritten."