Speculation: they could help v12 learn from v11 if both can run in parallel. If they trust v11 and have v12 run in shadow mode, any significant difference between the two could trigger a video upload to have the (auto) labeling team study the scene and have v12 learn from the observed success/failure.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
FSD discussion
- Thread starter Buckminster
- Start date
scottf200
Well-Known Member
Knightshade
Well-Known Member
Keep in mind there's hundreds of thousands of Teslas on the road with NO driver monitoring at all other than the steering wheel torque sensor (and millions more with an interior camera that was never designed for the job and can't deal well with sunglasses or darkness)
willow_hiller
Well-Known Member
Since the last firmware update, it's possible to go into the service menu and preview the interior camera. I have a 2019 SR+ with the non-IR cabin camera, and I was surprised how well it could see in the dark. Just the ambient glow of the center screen is sufficient to fully see my eyes and face.
I haven't tested sunglasses, but I imagine it's highly dependent on how dark/reflective they are.
jeewee3000
Active Member
I think the two most likely reasons are:
1/ during coding/improving/optimizing v12 they gain new insights. If these are relatively easy to implement in v11 to improve the v11 architecture, this is low hanging fruit. (For example the new efficient video module. That is the main improvement that translates into higher precision on many levels)
2/ They don't want to put all their eggs in the v12 basket. It could turn out to be the case that the v12 approach (E2E) requires bigger NN's/more compute than they can possibly provide/etc. If this is the case (in which HW4, 5 or more would be necessary), Tesla won't want to have an FSD stack (v11) being stagnant for years. To keep FSD customers happy, they have to throw breadcrumbs in the form of improvements at (somewhat) regular intervals. (If not, the take rate for FSD would drop also).
jeewee3000
Active Member
Indeed. The cabin camera can see more than you think. And that is only what we (as humans) can see on a screen. The raw photon input is even more detailed for the car.
Also, even with the darkest sunglasses on the market, it wouldn't pose a problem since drowsiness algorithms not only look for eyes, but also for body posture/head tilt etc.
When asleep or nodding off your head/body moves in a different position. This is obvious to a machine learning algorithm.
When deliberately doing something inattentive (for example watching a movie on your phone) you would have to move your head towards this distraction. Even with sunglasses this can be detected. (Unless you really try to game the system by mounting your phone on your windscreen, but any system can by cheated and this is the responsability of the driver).
willow_hiller
Well-Known Member
willow_hiller
Well-Known Member
Knightshade
Well-Known Member
I expect he'd just move to Microsoft like everyone else at OpenAI seems poised to do?
willow_hiller
Well-Known Member
That does seem like the most likely outcome. But the follow means Elon is interested in Karpathy's next move, at least.
mars_or_bust
Member
Actually, he might not. Karpathy is a very stand-up, nice guy, and it seems, reading a lot on the matter, that Sam Altman turns out to be a real nasty and manipulative kind of guy, maybe psychopath level. Such people are hard to at first identify as such if very intelligent, which Sam obviously is. But after a while the truth comes out. I would not be surprised if Sam actually is as described above, and that Andrey is slowly realizing that. If so, Andrey would not want to work with Sam. Which leaves X.ai or Tesla as the only other serious global players in town regarding AI.
Knightshade
Well-Known Member
Uh...what? Karpathy, like Sam, is one of the founders of OpenAI. It's not like he just met the dude last week.. Karpathy worked for him for years before Tesla and chose to go BACK to working for him after Tesla.
Not to mention like 95% of the ~700 employees all signed a letter not only in support of him but saying they'd quit and follow him to another company
So either what you've "read" about Sam ain't true, or Karpathy and 95% of all employees still haven't noticed.
Citation:
Introducing OpenAI
OpenAI is a non-profit artificial intelligence research company. Our goal is to advance digital intelligence in the way that is most likely to benefit humanity as a whole, unconstrained by a need to generate financial return. Since our research is free from financial obligations, we can better...
openai.com
That's from 2015.
OpenAI founding doc in 2015 said:
mars_or_bust
Member
Uh...what? Karpathy, like Sam, is one of the founders of OpenAI. It's not like he just met the dude last week.. Karpathy worked for him for years before Tesla and chose to go BACK to working for him after Tesla.
Not to mention like 95% of the ~700 employees all signed a letter not only in support of him but saying they'd quit and follow him to another company
So either what you've "read" about Sam ain't true, or Karpathy and 95% of all employees still haven't noticed.
Citation:
Introducing OpenAI
OpenAI is a non-profit artificial intelligence research company. Our goal is to advance digital intelligence in the way that is most likely to benefit humanity as a whole, unconstrained by a need to generate financial return. Since our research is free from financial obligations, we can better...openai.com
That's from 2015.
Intelligent psychopaths are very difficult to spot, especially by people on the spectrum - which both Elon and Karpathy are. I of course do not know if Sam is a psychopath, but so many small things seem to point to that. I personally had experience with several psychopaths over many years, and a lot of similarities are there.
Anyway, time will tell. Or not.
Knightshade
Well-Known Member
Todays "Ok, maybe we ARE in a simulation" brought to you by....
The new chairman of the board at OpenAI will be... the dude who was Chairman of the board of Twitter prior to Elon buying it.
The new chairman of the board at OpenAI will be... the dude who was Chairman of the board of Twitter prior to Elon buying it.
V12 is extremely unlikely to be ready by early 2024. I am curious how well V11 would fare in China though.
Similar threads
