I believe Elon has mentioned a few times that roughly 6 billion miles is the number needed for a true autonomous system. Taking what we know from transformer based LLM systems, massive scales of data truly do continuously improve performance beyond what most experts would have predicted even a couple of years ago. I’m not well-versed in this field, but I imagine there is an analogy between the neural nets learning the nuanced long-range language relationships as data scales compared to the image neural nets for self-driving learning nuances of edge cases in the physical world.
Having said that, maintaining production and sales at the expense of today’s margins will only continue to add to Tesla’s FSD dataset, which I think is critically important to FSD becoming a reality. In the most recent earnings deck, they showed a cumulative FSD miles plot that’s near exponential–I roughly hand-fitted an exp(0.19x) function to the data they showed. Extrapolating out that function would give over 500 million miles by the end of the year and around 6 billion by the end of 2024.
I know Elon has said FSD is coming for years, but he may become the broken clock that’s right in the next year or two assuming they can continue exponentially scaling their data.