Hacker News Reader

Comment by redox99 | Hacker News Reader

As someone who has been following Tesla FSD progress through the years through some YouTubers[1], the progress has been very very impressive.

2 years ago, FSD was pretty awful. Back then, my opinion was that Tesla FSD was doomed: Self driving without LIDAR was not going to be feasible for at least a decade. Now, I've changed my mind. Current version of course has a long way to go, but the current state and the rate of progress makes me believe now that, in 2 or 3 years, with a new hardware update, Tesla FSD will be very successful.

I recommend checking out some videos of the latest version (10.69.3.1) to get an idea of how good (or bad) it currently is.

[1] https://www.youtube.com/@ChuckCook https://www.youtube.com/@AIDRIVR

View on HN

"the rate of progress makes me believe now that, in 2 or 3 years, with a new hardware update, Tesla FSD will be very successful."

Isn't that exactly what people were saying 2 or 3 years ago?!

There's a lot of people in the world, I'm sure some said that. I would recommend drawing your own conclusions based on it's current state, rather than focusing too much on what people say.

You can check what I was saying a 2 year and 1.5 years ago:

I want from 50% chance after Sep 2024 to 50% chance after Aug 2027.

Rate of improvement IMO seems to be around 2x better every 6 months. But there is still a long long way from driving millions of miles without disengagement to meet their internal benchmark.

https://www.metaculus.com/questions/11608/self-driving-taxis...

I went to the first channel (@ChuckCook) and the first video I clicked had the car trying to jump an open Drawbridge...

https://www.youtube.com/watch?v=zoKAX-F1t9U

I feel like "Safety Score" was the only thing keeping this "FSD" beta to somewhat responsible adults (e.g. risk of losing access). When that they've opened the floodgates, how long before people are reading a book or on their phone while their car "self drives?"

It has a long way to go for sure. But remember that these channels focus on the hard stuff (narrow unmarked roads, unprotected left turns, etc). Also maybe I wasn't clear, but by successful I mean that your average Tesla owner will use it in their every day, and will be able to handle their daily commute fully. But they'll still be behind the steering wheel. I don't expect that in 3 years you'll be able to sleep while it drives.

Here's something more resembling an average drive from that same channel https://www.youtube.com/watch?v=b_1NOiFKukg

My prediction on Metaculus from May 2021 was Aug 2027. Today I still think that they will need 4-5 more years.

They seem to be going at about 2x improvement rate every 6 months.

https://www.metaculus.com/questions/11608/self-driving-taxis...

My very rough guess is that they still need to make it some 10000x better.

log2(10000) ∗ 0.5 year = 6.64 years

The improvement needed also roughly matches Tesla internal expectations.

According to CJ Moore, Tesla’s director of autopilot software:

> The ratio of driver interaction would need to be in the magnitude of 1 or 2 million miles per driver interaction to move into higher levels of automation.

So, my current prediction is Aug 2027:

- 5% before July 2023

- 25% before Aug 2025

- 50% before Aug 2027

- 75% before May 2030

- 95% before Apr 2034

To be fair I don't think it will be robotaxi level in 2 or 3 years. I agree that robotaxi level is probably more like 2027.

In 2 or 3 years by successful I mean that your average Tesla owner will use it in their every day, and will be able to handle their daily commute fully. But they'll still be behind the steering wheel.

I'm really curious how it will work out for them. The conventional wisdom in self-driving space seems to be that humans are not reliable controllers of automated systems that need input once every day or so.

I'm quite sure that there will be a lot of people overestimating this system if it tries to kill you only once per 1000 kilometers or once per 2 weeks.

Right now it tries to kill you every other drive, so people really pay attention.

> 2 years ago, FSD was pretty awful.

Full perfect self driving was complete and ready for release in 2017

> in 2 or 3 years, with a new hardware update, Tesla FSD will be very successful.

Do you think that folks who bought FSD will get a hardware update, without extra cost?

It probably depends on how well current hardware performs. If it's good enough they might not (however if you have a newer car with newer HWU FSD will perform better of course).

Camera placement (A-pillar camera), better cameras, and faster processing would be an almost immediate improvement to current FSD without having to improve software.

Well.. "impressive" is not good enough for public road. "Impeccable" should be the bare minimum.

I don't understand this attitude. Shouldn't the requirement be that it's better than the alternative, i.e. human drivers? Human drivers certainly aren't impeccable.

Reply on news.ycombinator.com