Tesla's FSD v12 is making unprecedented advancements in self-driving technology, with a rate of improvement that far surpasses previous systems and sets them apart from other automotive manufacturers
Questions to inspire discussion
-
What is Tesla's FSD v12?
—Tesla's FSD v12 is a self-driving technology that has made significant advancements, surpassing previous versions and showing a rapid pace of improvement.
-
How does Tesla's FSD v12 work?
—Tesla's FSD v12 uses a neural network model driven by imitation learning, similar to how humans learn to drive, and does not have a deterministic view of a stoplight.
-
What sets Tesla's FSD v12 apart from other automotive manufacturers?
—Tesla's FSD v12 uses an end-to-end approach with efficient AI models, customized from open-source, leading to significant progress, and has a rate of improvement that is five to 10 times better per month.
-
How does Tesla collect and process data for FSD v12?
—Tesla constantly fine-tunes their autonomous system by collecting and analyzing data from their 5 million cars on the road, leading to exponential improvements in their models, and uses edge processing to handle the massive amount of data collected by the car.
-
What is the potential impact of Tesla's FSD v12 on the market?
—Tesla's FSD v12 could potentially lead to a 20% penetration rate and a $500 monthly fee for users, and the company is focusing on driving adoption of FSD by lowering the price and increasing penetration to gather more data and stay ahead in the market.
Key Insights
- 🚗 The last 1% of achieving full autonomy in self-driving cars is as difficult as the first 99% due to the nearly impossible task of coding for all corner cases.
- 📈 The rate of improvement and change on FSD V12 is staggering, matching or exceeding the capabilities of all past versions combined in a fraction of the time.
- 🤯 The radical decision by Tesla to throw out the whole thing and start afresh is so damn impressive.
- 🚀 The rate of improvement of Tesla's FSD v12 is 5 to 10 times better per month compared to the prior systems, which is an extremely important fact.
- 🚀 Tesla's ability to look into the future and plan for long-term advancements sets them apart from other automotive manufacturers.
- 📊 The amount of Shadow data collected by Tesla's 5 million cars over 10 years is so massive that it surpasses the storage capacity of every hyperscaler in the world combined.
- 📈 The constant autonomous process of fine-tuning models is the reason behind the exponential moments of improvement in Tesla's FSD v12.
- 🚗 No one has the fleet, existing data, and AutoFosk for edge cases like Tesla does, making it hard for competitors to catch up.
#Tesla #FSD #Robotaxi
XMentions: @Tesla @StevenMarkRyan @HabitatsDigital @herbertong @theJeffLutz @AlexR6 @TheJeffLutz @GoingBallistic5 @DrKnowitAll16
@elonmusk @TeslaBoomerMama
Clips
-
00:00 🚗 Tesla's FSD v12 has shifted to a neural network model driven by imitation learning, using video data from their best drivers, showing significant advancements and a rapid pace of improvement, impressing a billionaire and highlighting the shift away from a deterministic view of stoplights.
- Two billionaires discuss their thoughts on Tesla FSD V12, highlighting the shift to an end-to-end model driven by imitation learning and the skepticism surrounding it.
- Tesla has shifted to a neural network model using videos from their best drivers as input, which has a better chance of success for full autonomy compared to their previous code-based approach.
- Tesla's FSD v12 has made significant advancements, surpassing previous versions and showing a rapid pace of improvement.
- Tesla made a radical decision to start from scratch and upload a lot of video data for their Full Self-Driving v12, which impressed the speaker.
- Tesla's approach to Full Self-Driving is based on a neural network system similar to how humans learn to drive, rather than hard coding instructions, and does not have a deterministic view of a stoplight.
- The new Tesla FSD v12 model uses pixels and driver behavior to drive, instead of heuristics and code, allowing it to adapt to various corner cases.
-
08:40 🚀 Tesla's FSD v12 is a significant improvement using end-to-end AI models and open-source components, focused on improving data infrastructure and models for successful autonomy.
- Tesla's FSD v12 is a significant improvement over previous versions, with a rate of improvement that is five to 10 times better per month, and the speaker believes that Tesla would have eventually solved autonomy with the patchwork model, but it would have been inefficient and messy.
- Tesla's FSD v12 uses end-to-end approach with efficient AI models, customized from open-source, leading to significant progress.
- Neural networks have been evolving for a long time, and Tesla's Full Self-Driving v12 uses the same hardware as the LLMs for training and inference.
- The Tesla FSD v12 utilizes open-source modular components that have been developed over the last decade, bringing together the energy of engineers to create an extraordinary model.
- Tesla's FSD v12 is focused on improving data infrastructure and models, with the hardware and data coming together to make it successful.
-
14:10 🚗 Tesla's FSD v12 uses edge processing to handle massive data, leading to exponential improvements in their autonomous system and leaving no chance for competition.
- Tesla has the capacity to look into the future and make long-term investments in technology, while other automotive manufacturers are late to the party and not seeing the potential of electric vehicles.
- Autonomy in vehicles is now possible due to the availability of compute power and the capacity for vehicles to collect and upload massive amounts of data, which no other company can currently match.
- Tesla's FSD v12 uses edge processing to handle the massive amount of data collected by the car, with only 1% of the data making it back to Tesla.
- Filtering and collecting vast quantities of data is crucial in order to find the outlier moments that are needed to train the model.
- Tesla is constantly fine-tuning their autonomous system by collecting and analyzing data from their 5 million cars on the road, leading to exponential improvements in their models.
- Tesla's FSD v12 architecture leaves no chance for competition.
-
19:54 🚗 Tesla's FSD v12 uses human feedback to improve, has a data advantage over competitors, and could lead to the development of humanoid robots in the future.
- Tesla's Full Self-Driving v12 uses reinforcement learning from human feedback to record moments of disengagement and abrupt movements, which is fascinating and could potentially improve the system.
- Tesla's FSD v12 can prioritize relevant data, making it impossible for competitors to compete.
- No company can compete with Tesla's Full Self-Driving technology due to their extensive fleet and data, while other companies like Cruise and Waymo rely on safety nets and overcomplicate their solutions.
- Tesla's data from their Fleet of vehicles has significant value, and it is possible for other companies to license FSD from Tesla directly.
- Tesla's FSD technology has the potential to translate into the development of humanoid robots, and the company's large fleet of vehicles could lead to a similar conversation about humanoid robots in the future.
-
25:05 🚗 Tesla's Full Self-Driving v12 is impressive due to its large quantity and quality of data, giving them a network and data advantage over other companies.
- The importance of having a large quantity of data, particularly millions of cars, is emphasized for the development of longtail events in Tesla's Full Self-Driving v12.
- Tesla's FSD v12 is impressive due to the quality and quantity of data it collects, but other companies may struggle to replicate it due to infrastructure and design control issues.
- Tesla has a network and data advantage, and with the improvement of their FSD, they could potentially reduce the price to increase penetration.
-
28:16 🚗 Tesla's FSD v12 could lead to a 20% penetration rate with a $500 monthly fee, focusing on driving adoption and having a more efficient business model than competitors.
- Tesla's Full Self-Driving v12 could potentially lead to a 20% penetration rate and a $500 monthly fee for users.
- Tesla is focusing on driving adoption of FSD by lowering the price and increasing penetration to gather more data and stay ahead in the market.
- Tesla's business model for autonomous vehicles is vastly different and more efficient than that of other companies like Cruz and Waymo.
-------------------------------------
Duration: 0:36:32
Publication Date: 2024-05-07T10:16:45Z
WatchUrl: https://www.youtube.com/watch?v=R3S9PH_VPFA
-------------------------------------
David Butler
Not sure what you and Elon have been smoking, but I’ve been on FSD since 2018, and while the city driving is impressive, assuming you don’t really care how long it takes to get somewhere, v12 on the highway is decidedly inferior to v11. My interventions are WAY up.
v12 doesn’t keep your speed up, recognizes route signs as speed limit signs, “keep your eyes on the road” is way too sensitive, windshield wiper settings are worse, it tries to pass cars on double yellow on two-lane roads, “Summon” is way worse than it was two years ago, FSD degrades in even the slightest inclement weather (likely due to the lack of radar)…I could go on and on, but the fact is that Level 5 is still many years away. There are far too many edge cases to make it realistic any time soon.