I predict that by 30-Sep-2025 it will be below 1000 vehicles and it will be under 10 cities.
1 Like
I predict that by end of Sep they’ll have between 1 and 3 cities and no more than 300 robotaxis in total, certainly not 500.
Faster scaling could happen in a year, but for now I think they will be getting lots of minor issues reported, trying to replicate them, retraining, running regression tests and put out OTA updates. It is doubtful they can repeat this cycle more than every 2-3 weeks.
They can probably get updates out to phone apps much faster than that.
But those issues aren’t generally safety critical.
Mike
3 Likes
The first 80% of the project takes 80% of the time.
The last 20% takes the other 80% of the time.
4 Likes
Adding to predictions thread.
We have some data on Musk’s prediction that Tesla will grow sales 20-30% this year.
Q2 sales are down 13.5% over last year, with deliveries of about 380,000 units.
Tesla doesn’t break out Cybertruck sales specifically, but it appears about 5,000 were delivered.
1 Like
I think Waymo tested in Austin with a safety driver (or employee on board, starting March 6 2024) for 1 year before going autonomous with general public riders (March 4 2025).
This calls for a prediction:
“By June 21, 2026, Tesla will not yet have
in Austin,
50 autonomous robotaxis (no employee on board = unsupervised),
driving the general public,
for fares,
in a geofenced area of at least 35 square miles.”
1 Like
Which approach will scale faster in the coming months?
Another prediction, I will take the other side (to keep all AI gods in balance and harmony and appeased).
Specifically,
“By June 30, 2027 (in about 24 months), Waymo will have
- more cumulative (lifetime) robotaxi miles
- more weekly (run rate) robotaxi miles
- more publicly disclosed robotaxi miles with incident data
than Tesla.”
Three predictions.
Who else has a prediction?
Fine print: Robotaxi means no on board employee, AI is in control (no local or remote control unless requested by AI, human phone-a-friend like Waymo protocol is allowed), available to general public, fares collected
1 Like
I predict that you just started a fight with the Tesla cult.
1 Like
Let this be a lesson in not taking one’s self, and certainly not any thread like this, too seriously.
You are good at that.
Would you serve as moderator?
1 Like
I predict your prediction will be wrong, but mostly because Tesla has not cared much about safety protocols or truth in advertising or other conventional norms. Further they will be under extraordinary pressure from the CEO and board to demonstrate speed and scale, perhaps before they are really ready to do so.
2 Likes
Thank you for your prediction and additional viewpoint.
I like the additional point of view, makes me feel so “inside the box” to be thinking about regulations and safety.
1 Like
How much data do you need to assess autonomous vehicle safety?
You need at least several million miles just to start analyzing safety.
To really get going, you need 10s of millions of miles.
Waymo has published incident (any-injury, airbag, or serious injury) data from 56.7 million miles over all locations and observed 68 incidents, so roughly 1 incident per million miles.
Waymo is currently accruing about 2 million miles per week from 1500 robotaxis across several locations (and in the process of adding locations).
1 Like
Quick update on this one. Musk tweeted that robotaxis would be in San Francisco in a month or two..
But California regulators report Tesla has not yet applied for the necessary permits, and declined to speculate how long the approval process might take.
Tesla has yet to apply for regulatory permits it needs to operate driverless taxis in California, two state regulators said on Thursday, a day after CEO Elon Musk said the company would expand its robotaxis to the San Francisco Bay Area within two months.
“To date, Tesla has not applied for either a driverless testing or deployment permit,” a spokesperson for California’s Department of Motor Vehicles said in an email to Reuters on Thursday.
1 Like
Even with all CA approvals today, Tesla will not have a robotaxi business in CA with any scale by year end.
Without some step-change breakthrough, Tesla has a long road ahead.
And the indications are that Tesla’s approach does not (maybe yet) have such a breakthrough.
What are the indications of no breakthrough?
- Calibrating cameras with lidar in a location-specific way (driving around Austin)
- Just now starting field service with safety drivers (and way behind Waymo in robotaxi miles)
- supposedly they have a data advantage - where is it? how and when will this advantage manifest itself? (looks to me like Waymo is the one with the data advantage of 56.7 million rider-only miles growing at 2 million per week and accelerating, Tesla has revealed 0 such miles)
So, brace yourself for the continuing and never-ending drip-feed of (relatively) minor features (“Grok, go to roadtrip playlist”) and teases and demos and projected timelines of vague promises.
1 Like
I have to ask, is any of this going to be cheaper than human labor?
2 Likes
Tesla was in the news because it just expanded the Austin geofence of its 10-vehicle, human-supervised robotaxi fleet that has restricted public access (Per Tesla: “Currently, Robotaxi is invite-only.”).
Summarizing my predictions from upthread.
Prediction 1
Here I mean: human-supervised is allowed (onboard or remote safety monitor/driver is allowed), must have non-employee riders paying fares (can still be very restrictive invite-only or more generally publicly available)
Prediction 2
Here I mean: human-supervised is not allowed (like L4 autonomy: onboard or remote safety driver not allowed), must be generally publicly available, must collect fares, must be in Austin metro with geofence at least 35 square miles
Prediction 3a,b,c
Here I mean: human-supervised is not allowed (like L4 autonomy: onboard or remote safety driver not allowed), must be generally publicly available, must collect fares, must be in US
1 Like
More predictions.
Who else has a prediction?
(Hint: consider the CEO’s prediction and predict lesser scope per cost: slower delivery, less functional product, higher cost)
If Tesla can’t even call FSD, without using Supervised, in the 10K than I think we can state that it is only L2 and everything else is a dream.
Further improvements and deployment of our FSD (Supervised) capabilities
Page 3 latest 10K just key on supervised.
https://www.sec.gov/Archives/edgar/data/1318605/000110465925042659/tm252787d2_10ka.htm
3 Likes
Tesla gave some new predictions on their Q2 earnings call on July 23.
Summarizing my predictions from upthread and adding some new predictions
Prediction 1
Here I mean: human-supervised is allowed (onboard or remote safety monitor/driver is allowed), must have non-employee riders paying fares (can still be very restrictive invite-only or more generally publicly available)
Prediction 2
Here I mean: human-supervised is not allowed (like L4 autonomy: onboard or remote safety driver not allowed), must be generally publicly available, must collect fares, must be in Austin metro with geofence at least 35 square miles
Prediction 3a,b,c
Here I mean: human-supervised is not allowed (like L4 autonomy: onboard or remote safety driver not allowed), must be generally publicly available, must collect fares, must be in US
Prediction 4
On the Q2 earnings call, Tesla said:
No, Tesla won’t develop autonomous (unsupervised) ride-hailing that is available to half the US population by the end of 2025.
human-supervised is not allowed (like L4 autonomy: onboard or remote safety driver not allowed), must be generally publicly available, must collect fares
Prediction 5
On the Q2 earnings call, Tesla said:
No, Tesla won’t develop autonomous (unsupervised) driving capability that is available for consumer purchase in any reasonably-size geography in the US by the end of 2025.
human-supervised is not allowed (like L4 autonomy: onboard or remote safety driver not allowed), must be generally publicly available for purchase by retail customers, must be capable in at least one geofenced area of 50 contiguous square miles
Maybe if Tesla buys the AI driving capability from another company, maybe then they might achieve some of the above autonomy goals?
Here’s a prediction from Sep 2017, 8 years ago.
The software has certainly passed the hardware, because the hardware is still changing to keep up with increasing software size (larger AI model).
You’re a true innovator when you scale production of unproven hardware.
And then scale it again for each new version of unproven hardware.
Tesla is definitely not “truly autonomous” on most local streets.
Not in 2020 and not in 2025.
Also not in 2026.
1 Like