Hacker News Reader

Comment by andrewtbham | Hacker News Reader

If you're interested in the facts...

https://www.tesla.com/VehicleSafetyReport

https://hcai.mit.edu/human-side-of-tesla-autopilot/

I'm not sure I would assume Tesla's POV here is necessarily "the facts" or that their safety site really informs much about the article.

Ok I added research from MIT.

Tesla has the data about it's fleet. Some guy that thinks he saw someone not paying attention is an anecdote.

Are you disputing their data? That's a pretty serious allegation for a publicly traded company.

>Are you disputing their data? That's a pretty serious allegation for a publicly traded company.

I'm skeptical considering Tesla's past approach to testing, their problems with the NTSB.

If you take as gospel every public corporation's PR materials, you should get in touch with Herbalife; I'm sure they'd love to sell you on a garage full of overpriced placebos.

I'm interested in this stuff, but their "quarterly reports" are just ~3 sentence summaries.

Is this the extent of these reports? (maybe I'm missing something). It would be helpful to see a list of the actual incidents they're using to calculate these numbers.

Sure, autopilot does better than the average car on the road. But there are parameters to control for. For example, the average car on the road is 10 years old, it's probably not very well maintained, and on average costs <$20k.

You should compare the safety of a Tesla to the safety of a similarly aged, similarly priced, and similarly equipped car, for example one with active safety mechanisms.

Plus, IMO the false confidence that Autopilot gives enables dangerous negligent behaviour from the drivers.

Want to see the safest car? NHTSA publishes an annual driver fatality report, and there are car models that have never had a fatal crash. See the 2009-2014 XC90 for example.

As mentioned several times on HN the comparison Tesla cites here is highly problematic.

There are no statistics that aren't problematic.

Do you have a different stat in mind you think would be more fair? Such as controlling for city/highway miles?

Probably every stat that isn't on the vendors marketing page.

Controlling for city/highway miles would be a great start. Then it would also be helpful for Tesla to compare itself against vehicles of a similar fleet age, price range (demographics make a big difference in the likelihood of a crash), and with similar modern safety features such as AEB.

Of course, they are famously cagey with all of this data, always preferring instead to point to the "Autopilot miles vs. all miles driven" number.

> “It’s obviously a misrepresentation,” said Hemant Bhargava, a UC Davis professor who specializes in technology management. “You’re only in autonomous mode in the best scenarios, so the number of crashes will be lower.”

Source: https://www.theinformation.com/articles/studies-dont-support...?

Reply on news.ycombinator.com