Hacker News Reader

Comment by AlotOfReading | Hacker News Reader

There are lots of cool things that probably shouldn't used by people operating heavy machinery. Phones for example.

From my experience working in this space though, judging quality is extremely difficult. For one thing, different people have different criteria. I've taken rides with other people where we all had different opinions on the quality of the ride at the end. These systems (FSD especially), can also be highly sensitive to specific details of the situation. Two drives in apparently similar conditions have meaningfully different behavior because the system understands them differently. It's very difficult to compare apples to apples outside statistics and simulation.

All this is to say: don't dismiss feelings or intuition about the danger. These are good caution signs and talking about them can help companies improve the product.

View on HN

I think everyone using this product should be asking themselves why we don't have an objective measure for the quality (or lack thereof!) of the product.

With every new release of FSD, there are people quick to say it's a breakthrough, and amazing, while others lament it doesn't nearly do what it says on the lid.

Why are we putting safety critical systems we can’t adequately measure on our streets?! There are people who confidently state that Tesla's FSD is already better than humans.

Where's the data?! Why isn't it open? Why can't we build an objective measure for effectiveness and safety of each of these releases?

Now this is an opinion I can get behind.

I'd love to see all cars (including Waymo) compete on a standard set of tests with clear gates as to what conditions they'd be allowed to go to Levels 3, 4 or 5.

The data isn't open because it's not required to be. Every manufacturer is either making a good faith effort to collect that data (to their own standards) or they probably shouldn't be on the road. Regulators, NHTSA especially, don't have the expertise, time, or mandate to get it from them, analyze it, and make it available. That could (and should) change though.

As an unrelated aside, measuring safety performance is hard. Doing it accurately without putting vehicles on the street is probably impossible. Simulation is not a fully adequate replacement for road testing and many companies are already looking at cutting the latter because of the expense. It'd be nice for regulators to outline an acceptable road test framework that better balances the goal of public safety from unsafe vehicles against ensuring safety can be demonstrated in real world environments.

Actually, there are some attempt to measure it.

https://www.dmv.ca.gov/portal/vehicle-industry-services/auto...

Of course Tesla is not participating - the real numbers would make them look disastrous.

Regarding objective measures for the quality - the famous standards for level 1-5 etc. Objectively speaking, Tesla's system is Level 2. There are other companies at level 3, and some others operating fully autonomous vehicles.

As to why Tesla chooses to base all their claims on purely subjective and highly cherry-picked numbers and videos ? We all know the answer to this question.

> Tesla's system is Level 2

Which I'm happy with. I do 95% of driving on Autopilot already. Add another 4% for FSD that I'm happy to babysit and I'd be very happy.

For some reason people are so caught up on naming choice and what it implies...

There are several reasons - for some it's about security, for other about truth.

I find it fascinating that a guy become the richest person in the world, by convincing 'investors' that a Level 2 system is 'almost ready' for a Robotaxi operation and will make them infinite money. To give him credit, he also did mention that Tesla is worthless if it doesn't solve FSD.

> There are lots of cool things that probably shouldn't used by people operating heavy machinery. Phones for example.

I will tell you straight up that a Tesla model Y on FSD 12.3.3 is a safer vehicle operator than a human being using a phone. Period. No qualifications. Under all circumstances.

Yeah, and? Being safer than an objectively dangerous behavior doesn't make it safe.

FWIW, I'm happy to see the "Yeah" there. But regarding the "and", it gets to the sincerity of the discussion. In a world where literally millions of people are texting from the drivers seat of a moving vehicle every day, freaking out on the internet about a handful of robots doing less dangerous stuff seems maybe a little performative.

I have a bit of professional interest here because building these kinds of systems safely is my job, whereas distracted driving isn't something I can meaningfully change as an individual. At the very least, I can explain to others what a safe development process looks like.

Reply on news.ycombinator.com