Tesla's AI Supercomputer

Discussions of the Dojo supercomputer, Elon Musk's latest obsession, have got me thinking: Might this be what finally unlocks true self-driving cars or makes for just another grand promise by Tesla's CEO? I've followed the developments very closely, and I'm not totally convinced.

Dojo is supposed to be this crazy AI training system that processes a huge amount of video data from Tesla's cars to update their Full Self-Driving capabilities. That all sounds very good on paper, but I have a lingering thought in my head: Have we not thrown enough data at the problem? There definitely comes a point at which more data doesn't matter.

What really interests me is Tesla's decision to develop its own chips for Dojo.

One way it can give them an advantage and not be dependent on pricey Nvidia GPUs. On the second thought, it is a very huge move that can also backfire hard if they can't match or outperform Nvidia. I have doubts as to whether Tesla can seriously compete in the chip game against established players.

These, to me, sound pretty ambitious timelines for the development of Dojo. Musk claimed that by February 2024 it would have been one of the top five supercomputers, but I have not seen anything to prove that this actually will happen. This indeed questions the realism of their other projections.

What might be interesting, though, is the potential applications of Dojo beyond FSD. Training general AI with it or renting out compute power could be a source of new revenue streams for Tesla. But I am not sure if the incompatibility of the software makes this a near-term viable solution.

What really concerns me is the massive investment Tesla is pouring into Dojo. We are talking billions of dollars here. Ambition is good, but I'm afraid it might turn out to be a resource vacuum for other areas in a very important manner.

I'm not entirely sold on Tesla's camera-only approach to autonomous driving, either. Most all of the other firms in the space utilize a bunch of sensors, and I just can't help but think that there must be some reason for it. Is Tesla being held back by it's "stubbornness" on this point?

The recent news about Tesla still pounding a lot on Nvidia hardware for AI training makes me wonder how far along Dojo really is. Seems like hedging their bets, which is probably smart, but sure doesn't inspire confidence in Dojo's readiness.

It's going to be interesting to see how all of this comes into play. If Tesla manages to get Dojo out and actually has full self-driving capability, that changes everything. I've just been burned by one too many of Elon Musk's overzealous timelines to hold my breath.

Posted Using InLeo Alpha