Transformers won because they allowed data to flow like water within the model.
- Transformers won because they allowed data to flow like water within the model.
- Don't be clever about where it flows, just let it flow everywhere.
- Just increase the bandwidth from the past to the future.
- No one has done that at the social level yet (without being a hyper aggregator).
- The transformer architecture has extremely high leverage.
- It's only a few hundred lines of code.
- What if you could have the same kind of runtime for data in general?
- A small set of physics and policies that allow data and code to flow like water.[hn]
- What would that unlock?