I don't think the models are the bottleneck. Instead, how and where we deploy them remains significantly underrated and underutilized.

Like imagine how crazy that you can get human-like intelligence in a small device, we should be able to do more than a chat interface.