The moat right now in #ai is the model, the moat in a year will be memory (how well the system remembers YOU and your queries and picks the right prompt and model, aka we are headed back to recsys real soon.)
There’s a reason ChatGPT started storing user session data although it’s risky and annoying from an engineering and legal perspective, and why Meta is putting semantic and hybrid search in all its products even though it is orthogonal with a good search experience in those interfaces.
