The speed with which Waymo is standing up driverless operations in San Francisco and LA definitely disproves the theory that they would be stuck in "easy mode" because they started in Phoenix.
The only problem is that the operations they are scaling aren't profitable. That's the hurdle that counts in robotaxis now.
@niedermeyer I have argued that instead of terms like "solving" or "achieving" (which is the typical nomenclature that Tesla employs for #FSDBeta) for Level 4-capable fleets that the true metric is a "Passenger Revenue-Minus-Continuous Validation Economic Model" - which closely resembles commercial air travel.
Continuous validation is extremely costly and continuous validation never ends (hence, why a concept of a "finish line" in safety-critical systems does not exist).
@adamjcook I think that's a really reasonable approach.
Lately, I've been telling people about how my dad let me drive at age 12 on camping trips in Eastern Oregon, taking over when we would get to towns. There's no question I could drive, I just couldn't get a license, insurance, a job driving, etc. That's all part of the driving value chain, and what needs to be solved for.
@niedermeyer As you know, the horse has been beaten to death, but the hard obligation of continuous validation is why this concept of "generalized self-driving cars" is so unrealistic.
It simply cannot be validated - initially or continuously.
And if it cannot be validated, the business risk is forever unquantifiable.
There is *far* more to a safety-critical system than just "an AI".
@adamjcook @niedermeyer hang on, did someone mention AI and safety critical? 👋
🤣🤣🤣
As an aside, Waze and Cruise aren’t really achieving the whole “driving” bit really either. They’re generally making it around but not well, not consistently and always trying to avoid scrutiny. Forget making money and get it to work (which it never will cause of how they’ve chosen to do it).
@CrackedWindscreen @niedermeyer It is going to take years to play out - one way or another.
And Waymo and Cruise (amongst others) are undoubtedly going to have to dump *enormous* sums of money into their continuous validation efforts... even in "peace time".
Then, inevitably, incidents will happen - which then requires a brutally-quick "root cause analysis-to-revalidation" process industry-wide.
You are absolutely correct on the lack of regulatory scrutiny. That will not last.
@CrackedWindscreen @niedermeyer I think... the way that I would define the industry's approach now is one which seeks to avoid a collision with a vehicle or a human *at any cost*.
It is a reasonable approach on paper, but in practice, it puts a myriad of very dangerous *indirect* incidents on the table - incidents that naturally fly under the radar.
We have to assume that these incident types are occurring, are causing harm and are going undetected.
E.W. Niedermeyer
Adam Cook
Andrew Clews