ImageNet and the Turning Point

Some turning points are only obvious after the numbers come in.

By the end of the last chapter, the three missing ingredients were finally in place: the algorithm, the hardware, and the data. This chapter is the moment they came together in one room, in one result, and the field changed direction almost overnight.

Progress in science usually moves slowly, then suddenly. Years of small improvement, error rates nudging down one percentage point at a time. And then something arrives that doesn't fit the curve. A result so far outside the expected range that it doesn't just move the conversation forward. It ends one conversation and starts another. People who had been working on something else look up. The field reorients.

That kind of moment is rare. When it happens, it compresses years of debate into a few months of consensus. The question stops being whether something works, and becomes how fast to catch up.