NN is supposed to mimic "General Intelligence" which is a misnomer according to the father of NN (see the AI podcast). You don't need "AGI" for driving as such and when they talk about AGI in the chat gpt context, they don't include driving, anyway.
It’s not like I missed GPT-1 or GPT-2 before GPT-3 and then ChatGPT. Nor did I miss AlexNet, AlphaGo, AlphaZero or AlphaFold. But hey, welcome to the field!
The framerate of other vehicle visualizations from the v12 livestream and 12.1 videos seem to be on average closer to 10 frames per second whereas 11.4.9 videos seem to be closer to 20 fps. At least earlier 12.x versions seem to be described as running end-to-end control at the maximum 36fps matching the camera inputs, so potentially the existing neural networks are kept around to periodically update the visualization. If the existing perception is reused as inputs to end-to-end control, it could be running at higher framerate for decisions but lower for visualizations. Or it could just be purely used as a visualization aid to provide some context for the blue control path visualization but not directly used for control.
That's a bad analogy, you are talking about a clear device failure (on a device assumed to be 100% accurate when working) or determinable noise on only one, not a qualitative threshold decision.
The neural networks probably want to be trained on diverse data from around the world so that it can hopefully understand the potential behaviors of other vehicles. These might not be specific perception outputs of existing 11.x, but if 12.x networks internally can learn these differences, it might even lead to control behaving differently based on what it sees other vehicles doing. Similar to how large language models will complete a sentence differently based on the subject, e.g., Sally vs Bob vs Fido, and how some assistants are biased to be less biased, end-to-end control might actually be quite sensitive to perceived context and could require debiasing to say follow further at 3 seconds even if everybody else is tailgating.
It's actually appropriate. Unless one of the two stacks crashes or has a detectable failure, how do you determine which of the two is wrong? Maybe they are both acceptable, in different ways. Or, maybe they are both wrong.
You are talking about Segal's law which talks about the problem of having too many or too little information. The man who has a single watch does not in fact know if his time is accurate and the man with two watches does not know which one is more accurate.
Or, you have two people taking notes. One writes, "The CEO wants donuts, not bagels, at the next meeting. The other writes, "The CEO wants bagels, not donuts at the next meeting" You weren't at the meeting but are responsible for refreshments. Which is correct? Should you start looking for a new job now?
Which is why Cybertruck's steer by wire has three steering wheel sensors and dual rack position sensors with different measurement methods.
I’m sorry that’s too many information, how will they know which one is more accurate. They should use only one. Lol
I wasn't necessarily referring to your specific meetings. Only setting a hypothetical where the output is processed data, not a recording straight from a sensor. BTW, there are many workplaces where audio/video recording is forbidden.
Just about the only thing I love about this company (bar management) as a tech enthusiast is the willingness to move fast break things and keeping on iterating and willingness to take risks. Can’t wait to see the progress 2024 holds.
