You just flip flopped.
Before you say there's not proof that more data = better model.
Now you are saying not proof that more data = signficantly better model.
The correct answer is, more (clean) data will always improve the model, but usually with increasingly diminishing returns.
BTW, that's why model size has to go up. You need both to get the best models.
Which is why FSD will rely on bigger and bigger models over time. HW5 should work I hope.
Agreed on all counts, I should've been clearer in my initial message.
What I meant is that there is a point at which more data isn't going to make your autonomous driving system better than others. From what I'm seeing Waymo already has enough data to achieve it in a way that won't affect commercial viability for RT. Sure, Tesla's might end up being autonomous everywhere, but as many other posters said before me, RT business is one where the majority of revenue is done in a relatively limited area. You cover the big cities and their outskirts and you're done. You might argue that, beyond having a subscription business (similar to FSD now), it won't be commercially viable to have autonomous vehicles go too far from their base.