Pouring more computational resources—bigger models, more data, longer training runs—into AI training tends to yield predictable capability gains. This "more compute in, more capability out" empirical pattern has been the single biggest driver of AI progress over the past decade.