Cognition Founder: AI has surpassed humans in pure reasoning; our last advantage is memory retrieval.

AIMPACT News, May 11 (UTC+8), according to Beating Monitoring, Colossus Magazine published a three-hour exclusive interview with Cognition founder Scott Wu. In the interview, Scott Wu discussed a rarely considered question: what exactly makes humans stronger than AI? His answer is not reasoning—AI has already won in pure logical deduction. The true strength of humans lies in retrieval, the ability to “suddenly remember” something. He gave an example of a phone verification code. A two-step verification code is 6 digits; if it were 10 digits, most people probably couldn’t remember it. The human brain’s “context window” is that short. But humans have one thing that AI cannot do: when you open a code file, suddenly four months ago, you remember a related bug you fixed—what happened, how it was fixed, why it was changed—all flooding back at once. Wu calls this “soft context,” not word-for-word memory, but the experience you’ve accumulated over the years, precisely triggered by a certain clue. He said that every day, in Cognition, his decision-making relies not on deriving a logical problem from scratch, but on automatically recalling long-held implicit judgments like “why we designed it this way” or “after the new technology emerged, we decided to migrate to another framework.” AI’s context window is much larger than humans’, but it still falls far short in cross-time and cross-scenario associative retrieval. AI is catching up in two ways. One is vector retrieval (embedding search), pulling relevant segments from vast amounts of historical data, which directly mimics human associative memory. The other, more aggressive approach, is continual learning, allowing AI to update its weights while working, similar to neuroscience’s “neurons that fire together wire together.” Wu believes the second approach is closer to how the human brain actually works, but it is still in early stages. Near the end of the interview, he said: “Ultimately, doing AI is really about solving cognition itself. Humans have been accelerating everything for thousands of years, but one thing that hasn’t changed is that someone still needs to think and invent. Now, even that premise might be changing.” (Source: BlockBeats)