Bear case for Tesla: 1 sentence

[mostlylong]

@gledmo

IMO, I actually think mostlylong’s points may be somewhat conservative.

I agree, I believe they are conservative so I used phrase “at least” a few times.

On this one:

@gledmo

'Tesla needs at least 20 million L4 autonomous miles to start to statistically demonstrate human level safety."

Let me explain further.

Let’s distinguish between the number of miles to train a model versus, given a trained model, the number of miles to demonstrate (measure) safety at human level or better by that (already) trained model.

I am discussing the latter.

Here is a thread on that topic:

@mostlylong

To show, statistically, that an AI driver is 25% better than a human (incident rate is 75% of human rate), one power analysis (with 80% power) estimated that about 28 million driverless miles are needed when the incident is defined as “police reported accident.”

and the Waymo study on same:

@mostlylong

Waymo study (https://www.tandfonline.com/doi/full/10.1080/15389588.2024.2380522)

Finally,

@gledmo

“Tesla is at least one order of magnitude away from human or better safety levels”, in my opinion it is clearly more than just one order of magnitude. Is it as many as 3 or maybe even 4?

One challenging item is that the AI driver could be 3 or 4 nines good: 999 or 9,999 miles without an accident-preventing human intervention.

But this is still terrible: 1 accident per month or year, roughly (details depend on exact definition of accident severity, more severe accidents are rarer).

But to the average person driving FSD, this might seem really good: they only intervene 1 to a few times per year.

Of course, are people really letting FSD drive 100% of their driving miles? And any route they would normally take? Almost any environmental conditions/weather/time?

No.