CityNerd has hilarious dead-pan takedowns of Musk’s Vegas Loop and Hyperloop. As for Musk’s planned Nashville Loop: “The rock is harder than it should be.”
Musk sounded the alarm about the dangers of AI, but it turns out it was just part of his scheme to get control of OpenAI. He shut up about it after launching xAI.
> But the helicopter view is that, unlike other autonomous vehicles, Tesla’s system has zero redundancy or safety nets and requires a nearly 100% accurate AI — which categorically can’t exist — to be even remotely safe.
This is probably the best high level explanation I have seen for why Tesla’s approach to autonomous vehicles can’t work. I think most people understand by now that AI makes mistakes. There needs to be a fallback for when the AI fails, and Tesla doesn’t have such a fallback.
The reason we call it an AI is because it is a trained model, as opposed to being an algorithm (traditional computer program). As far as I am aware, trained AI models can’t be verified to be 100% accurate. It’s more a matter of trying to do the best job possible of measuring failures, and iteratively making that failure rate as small as possible. You’ll never reach zero though, with current technology.
Musk was never a Tony Stark. He is a Pakled.
CityNerd has hilarious dead-pan takedowns of Musk’s Vegas Loop and Hyperloop. As for Musk’s planned Nashville Loop: “The rock is harder than it should be.”
Musk sounded the alarm about the dangers of AI, but it turns out it was just part of his scheme to get control of OpenAI. He shut up about it after launching xAI.
> But the helicopter view is that, unlike other autonomous vehicles, Tesla’s system has zero redundancy or safety nets and requires a nearly 100% accurate AI — which categorically can’t exist — to be even remotely safe.
This is probably the best high level explanation I have seen for why Tesla’s approach to autonomous vehicles can’t work. I think most people understand by now that AI makes mistakes. There needs to be a fallback for when the AI fails, and Tesla doesn’t have such a fallback.
The reason we call it an AI is because it is a trained model, as opposed to being an algorithm (traditional computer program). As far as I am aware, trained AI models can’t be verified to be 100% accurate. It’s more a matter of trying to do the best job possible of measuring failures, and iteratively making that failure rate as small as possible. You’ll never reach zero though, with current technology.