Jamin, I wonder if open source may be inheriting more assumptions from the frontier labs than we realize.
The article focuses on research breakthroughs, model quality, and the widening gap between frontier and open models. Those are obviously important. But historically, technologies don't always gain traction because they become technically superior. Sometimes they gain traction because someone discovers the application that makes their value obvious.
The early web is an interesting example. By modern standards, PizzaNet was primitive. Yet it pointed almost directly at the future internet economy because people immediately understood what the technology was for.
It makes me wonder whether the biggest opportunity for open source is not necessarily to match the frontier models first, but to discover a commercially compelling use case that the frontier labs are overlooking. If that happened, the ecosystem and funding incentives could look very different very quickly.
Incentives rule. As Charlie Munger said: "Show me the incentives and I will show you the outcome." One potential dis-incentive here is the growing cost of compute. Without Moore's Law exponentially lowering costs for us, we now have to actually pay for the compute we use and that bill is getting big enough to dis-incentivize this type of open source risk taking.
Jamin, I wonder if open source may be inheriting more assumptions from the frontier labs than we realize.
The article focuses on research breakthroughs, model quality, and the widening gap between frontier and open models. Those are obviously important. But historically, technologies don't always gain traction because they become technically superior. Sometimes they gain traction because someone discovers the application that makes their value obvious.
The early web is an interesting example. By modern standards, PizzaNet was primitive. Yet it pointed almost directly at the future internet economy because people immediately understood what the technology was for.
It makes me wonder whether the biggest opportunity for open source is not necessarily to match the frontier models first, but to discover a commercially compelling use case that the frontier labs are overlooking. If that happened, the ecosystem and funding incentives could look very different very quickly.
Incentives rule. As Charlie Munger said: "Show me the incentives and I will show you the outcome." One potential dis-incentive here is the growing cost of compute. Without Moore's Law exponentially lowering costs for us, we now have to actually pay for the compute we use and that bill is getting big enough to dis-incentivize this type of open source risk taking.