[voramok]

I still believe that his insistence on only using computer vision for Tesla’s self driving was profoundly wrong. I got my ivy-league graduate degree in a related field right around the time when he probably was making the decision (2015) and it was obvious to everyone around me that we are not even close to having the kind of generic, reliable CV you would need for a car. Obviously, in grant proposals and papers we painted a much rosier picture, but on the inside everyone knew it’s just exaggerated, best-case scenarios that have little bearing on reality.

[indolering]

Also in a related field (cognitive and neural linguistics) and I 100% agree. Elon read too much science fiction and doesn't understand the full stack. The results from automated statistics (I hate calling it machine learning) is amazing - but it amounts to really good pattern matching.

Which is why Tesla Autopilot struggles with left turns and has been stuck at SAE level two after a decade of work. Hell, Tesla Inc failed to automate their factories much beyond what other automakers already accomplished and that is for single-purpose robots in fixed operating environments. Yet Elon is busy selling shareholders on humanoid Tesla Bots that will be capable of running to the grocery store and getting items off the shelf ... someday.

People with a background in the hard sciences just don't do well when grappling with the soft sciences .

[superchroma]

Considering I have been in situations with zero optical visibility, I am frankly baffled he thought that removing sensors including ultrasonics which are cheap as chips, was a good idea

[indolering]

It makes sense if you are someone who sold cars with a $10K "autopilot" add-on claiming that they would eventually be capable of "full self driving" for a decade only to realize that it's not feasible from a technological standpoint (at least not before most of those cars arrive at a junkyard [1]) AND the global supply chain suddenly shits the bed - jacking up the price of parts (especially electronics).

At that point, you can either (A) stop selling models that you actually make a profit from and risk your top spot on Forbe's billionaires list or (B) claim that "full self driving" really meant "driver assistance" this whole time [2] and cut whatever hardware isn't absolutely necessary to meet your lowered standard.

Elon chose option B - I suspect through some mix of self-delusion about the capabilities of automated statistics and knowing that he can continue to push the deadline for "full self driving" back another decade and pivot after his competitors have pushed the cost of radar/LIDAR down far enough.

[1]: https://www.nbcnews.com/id/wbna12040753

[2]: https://www.insidehook.com/article/vehicles/controversy-tesl...