Walk through the lobby of NeurIPS now and it barely resembles a "conference" in the old sense. You have LED walls and sponsorship towers. Recruiters handing out hoodies like it's a music festival. Private suites upstairs where the actual negotiations happen.

If you ask people how open-source AI is "governed," a lot of them will tell you it isn't. "No one is in charge." "Code is law." "Anyone can fork, that's the governance." That sounds romantic. It's also wrong.

If you hang around AI long enough, you see the same pattern on repeat. A flashy preprint drops. Everyone posts the same figures on social media. Repos and re-implementations explode for a month. Then, either the idea quietly dies, or five years later it shows up inside a boring enterprise system that never mentions the original paper.

For a few years, the field lived under a simple rule of thumb: if you make the model bigger, give it more data, and spend more compute, it gets better. The original scaling laws for language models turned this intuition into curves on a log–log plot. Loss fell in a smooth, almost comforting way as parameters, dataset size, and compute all went up together.