44.x is supposed to fix a vibration issue on Plaids...
2023.44 contains non-FSD software changes.
How does this type of stuff affect the OEDR? For example in the v12 clip that Elon posted to Twitter, his vehicle at one point exhibited the same red light issues that have existed for many versions (since inception likely) where the vehicle can sometimes see an adjacent green light controlling another lane, mistakenly assign the green light to its lane, and tries to proceed on what is actually a red light.
Classic NN training would mean adding training data to correct any problems. You can also “over sample” edge cases to emphasize them.
We don't know, which is why we're holding our breath while Tesla goes through the process. We don't know what the vehicle hardware is inherently capable of, and we don't know how much of that capability Tesla's engineers are capable of exploiting. If you're trying to get amped up in anticipation of it, forget it. There's no point in doing that now, just as there has never been a point in doing it in the past. Just wait and see what they pull off.
Holiday version will be 2023.x.25. That’s what they have done in the past. Could be 44 or 48.
The potential for end-to-end is more "automated" fixing of issues as any intervention can go through the streamlined / no-engineers-required process of auto-labeling and training. There can still be special focused effort to take on specific issues like running red lights or unprotected lefts, and this could be in the form of curating data collection for these scenarios, and even this can be relatively low engineering effort than before as 12.x's design hopefully is general enough to avoid code or neural network architectural changes.
I assume that the front camera is just a stand-in for ultrasonic sensors. That is, it isn't involved in the autonomy system. So you're still looking at HW4.
Or is not what you thought!
Yeah, this would likely be the case for the foundational / world model generally understanding how things behave based on the cameras and other inputs. So adding say an extra camera with "HW4.1" would probably go through a much longer process of collecting video then training for the model to internally understand how to stitch in a front bumper camera. Similarly, differences in camera resolution and positioning will likely need some extra effort if there's a unified base model shared across all vehicles.
Just making sure, you're suggesting that the end-to-end livestream from August is a separate development path from what we'll get with 12.x because 3 months from demo to employee testing is not enough time to make things safe? Tesla had been working on it with enough confidence in the approach to present at CVPR in June, and from Isaacson's biography, Elon Musk tried it out in April and was impressed enough to focus resources on this approach for over half a year now.
Who knows !?
This is definitely not my expectation- I think it’s a major rewrite and (mostly) end to end.
