Hacker News Reader

Comment by cma | Hacker News Reader

The much bigger news is they completely walked back FSD (full self-driving) into something that you have to monitor at all times by changing the description on their website.

The release of a proprietary charger is somehow completely overshadowing that news.

It's changed from driving itself around in a ride hailing network to earn you money while you sleep, to now making you always be in the car and pay attention and make corrections, essentially turning all purchasers into unpaid neural network training practitioners.

View on HN

I think you're confusing the next incremental step toward FSD with the end state of FSD. Tesla changes its marketing materials and web site copy all the time, but the Master Plan Pt 2 post (which outlines what you're alluding to) remains unchanged. Additionally, just three weeks ago, Elon was on the Ark Invest podcast claiming FSD will be safe enough for a human to sleep at the wheel by end of 2020[1]. I don't think Tesla's FSD plans have changed.

1: https://ark-invest.com/research/podcast/elon-musk-podcast

And four or five weeks ago Tesla had a filing about building more stores.

It seems a lot of plans are changing.

Elon claiming that in a podcast isn't a promise like this text on the order page.

According to the vehicle purchase agreement and Tesla's lawyers, nothing on the order page is binding anyway (and ditto for Elon's claims). I agree they have changed the "promise" they are making to new purchasers on the page you've cited, but I don't see evidence that they are scaling back what they intend to deliver vis a vis FSD.

Honestly, every time I read the "you're turning me into someone working to train your program without paying me!" comment I think about this stuff and I realize no one I know would give a goddamn fuck and expect to be paid in cash. Having the better product is sufficient.

These comments remind me of how much a bubble hacker news is. None of my friends would know what the hell a neural network was, and would laugh out loud if they heard that comment.

Precisely why I said in the past here that it was a big mistake to fit all model 3s with self-driving hardware making the car significantly more expensive in terms of cost to build.

Musk once again overestimated by a long shot how good and safe autopilot will be. At least it seems he is becoming more responsible or maybe he is forced to by the new board.

He recently publicly announced that Tesla drivers will be able to fall asleep at the wheel and wake up at the destination by the end of next year. Doesn't sound like a lot changed in his mindset.

Sounds like a good chance of waking up at the "final destination" or at least waking up having sent a pedestrian there.

He claims a lot things...

Yes, i don't give a rating to this statement, i just say that Musk is not conservative about this topic as op suggested. If you ask me, I don't think that they will be anything close to that by the end of 2020

Given that Tesla is considered one of the poorer self driving options out there, and the other players in the game say 5-10 years, maybe I need to hit up a bookie in Vegas...

>Precisely why I said in the past here that it was a big mistake to fit all model 3s with self-driving hardware making the car significantly more expensive in terms of cost to build.

Why? data, that's why. You need massive amounts of good quality data to build a solid model that will be 99.9999% accurate. This is one of the leverage Tesla has against Waymo.

You can simulate all you want but nothing beats real world data.

Link the change?

"Tesla updates self-driving language, creates confusion and walks back features"

https://electrek.co/2019/03/06/tesla-self-driving-language-w...

Old language:

https://i1.wp.com/electrek.co/wp-content/uploads/sites/3/201...

New language:

https://i1.wp.com/electrek.co/wp-content/uploads/sites/3/201...

All mention of the ride hailing service is gone, and now unsupervised driving isn't just dependent on just regulatory approval like they said before but now is instead dependent on it ever "acheiving reliability".

Just taking out the car-sharing bit? Meh.

> essentially turning all purchasers into unpaid neural network training practitioners.

Pretty sure customer data isn't uploaded en masse to train networks at Tesla.

They run in shadow mode to gain statistical data, and I believe when actively used they can send more detailed data around disengagements.

I bet of they allowed people to opt in, at least a huge minority of Tesla owners would jump through hoops (like login and recaptcha) to contribute data if it advances the science.

> By November 2016, Autopilot had operated actively on hardware version 1 vehicles for 300 million miles (500 million km) and 1.3 billion miles (2 billion km) in shadow mode.[21]

https://en.wikipedia.org/wiki/Tesla_Autopilot

Reply on news.ycombinator.com