I agree I've noticed I need to be quite specific now, it won't realize bugs in the code unless I tell it so. A head to head comparison of the different versions is needed to validate this.
Haven't read the paper, but it's still unclear how the mechanism of one shot learning works. If the weights are not being updated, how is it "learning"?
Tesla has Full Self Driving as a clear objective and feature though, autopilot is just their way of releasing the useful features they acquire along the way to FSD. Tesla's difference in its approach (which I think most people tend to miss) is that it aims to be a robust general solution to L5, not a "hard coded", expensive, brittle, geofenced solution, which are basically what all the other L4 startups are aiming for.
As famously Levandovski (the guy behind Waymo, Uber and Otto) and others have accepted that Lidar and HD maps are the wrong approach.
https://medium.com/pronto-ai/pronto-means-ready-e885bc8ec9e9
It just is not a sufficiently general and economic solution to be scalable across time and space. It doesn't matter if Waymo's only focus has been FSD (which it completely hasn't) if they have a wrong approach they won't achieve it.
Not really, Uber has accumulated around 3 million drivers, Tesla has around 1 million cars with FSD capabilities and growing much faster than Uber. Tesla has the advantage that their fleet would be composed with their owned cars plus customer cars, my prediction would be that they would quickly becoming bigger than Uber's fleet of drivers once they have FSD and with a click of a button you could send your car to make money for you.
Great job, ocean cleanup very interesting data right here. Tracking the problem in that way is more of a first principles approach. Happy that Mexico is so low on the amount.
I don't know about this. Countless times researchers discover the next battery so much better than the current state-of-the-art being deployed, only to find out they just can't scale it. New battery technology is really hard.