[kamranjon]

The part that I always found difficult to square was not that Elon Musk towed this “no need for LiDAR” line so hard, but that Andrej Karpathy, who I generally consider a very reliable voice in this space, was also in strong agreement that cameras were all that was needed. Does anyone know if he still believes cameras is all you need?

Edit: Here is a link to Karpathy discussing the trade-offs: https://www.youtube.com/watch?v=cdiD-9MMpb0&t=5276s

[chronic0262]

> but that Andrej Karpathy, who I generally consider a very reliable voice in this space, was also in strong agreement that cameras were all that was needed.

You really expect a Tesla employee to speak out against Elon?

Especially when $10M+ TC is on the line?

[rumblefrog]

But Andrej is no longer with Tesla?

[sgt101]

Andrej "train it on more data and the problem will go away" Karpathy

[Grimblewald]

Right? Im not arguing against the skills he obviously has, but if we're always just piece wise approximating the underlying manifold, then there will always be new problems. The amount of data required to reliably approxate reality, in the absence of an inductive bias, is infeasible to expect to collect. Not to mention how computationally inefficient it becomes as your model blows out in size/complexity.

[sgt101]

I guess the gamble was that there would be a certain point where the edge cases disappeared into the noise and it didn't matter that the approximation was/is "wrong" because the behavior of the car would match the requirements of the situation even if it wasn't for the right reasons.

To be fair I remember reading about GPT2 and thinking that LLMs would blow out for similar reasons.

[adastra22]

Which they more or less have. Larger models are seeing negligible returns. It just turned out that scaling would hold out just enough longer to make LLMs generally useful.

[Grimblewald]

Yup, and if you normalise "improvements vs time" graphs to not linear time but gpu hours invested per unit improvement we're in extremely incremental/small improvement territory as of a year ago. There are no major jumps coming. There are no more gpu hours to allocate to dumping onto this partciular bonfire to keep things looking like exponential improvement, all to keep that vc cash flowing.

[satyrun]

Isn't it ultimately a cost trade off? I mean I can't see a valid argument against LiDAR and cameras if the cost of the vehicle is no concern.

If building a mass market product though the cost is a big deal.

I would assume LiDAR is much more expensive so it would be a big win to get the same performance out of cameras in the long run. I have always just assumed that was the bet.