KuMX
Well-known member
- First Name
- Ku
- Joined
- Nov 15, 2019
- Threads
- 170
- Messages
- 421
- Reaction score
- 2,726
- Location
- Socal
- Vehicles
- Cybertruck AWD + 2017 Model X
- Thread starter
- #1
https://www.notateslaapp.com/news/2526/whats-coming-next-in-tesla-fsd-v14
So, what’s next with FSD V14? Tesla gave us a sneak peek at what’s next for FSD.
FSD V14
The standout feature of FSD V14 will be auto-regressive transformers. While that’s a complex term for those unfamiliar with AI or machine learning, we’ll break it down.
Auto-Regressive
An auto-regressive transformer processes sequential data in time, using that information to predict future elements based on previous ones. Imagine completing a sentence: You use the words already written to guess what comes next. This process isn't just about filling in the blank; it's about understanding the flow of the sentence and anticipating the speaker's intent.
FSD could analyze a sequence of camera images to identify pedestrians and predict their likely path based on their current movement and surrounding context. The system's auto-regressive nature allows it to learn from past sequences and improve its predictions over time, adapting to different driving scenarios.
Today, FSD reacts to what it sees, but soon it’ll be able to anticipate what will help, much like humans.
Transformers
The second part of that term is transformer, which is a component used to understand the relationships of elements inside a time sequence. It identifies which parts of the input are most crucial for making accurate predictions, allowing the system to prioritize information much like a human would. Think of it as weighing different pieces of evidence to arrive at a conclusion. For example, a transformer might recognize that a blinking turn signal is more important than the color of the car when predicting a lane change.
Putting It Together
Putting all that together, Tesla’s use of auto-regressive transformers means they’ll be working on how FSD can predict the plans and paths of the world around it. This will improve FSD’s already powerful perception and allow it to predict how other vehicles and vulnerable road users (VRUs) will behave.
What it all comes down to is that FSD will be able to make better decisions and plan its paths by making more informed, human-like decisions. That will be a big step towards improving V13 - which already has some very effective decision-making.
Larger Model and Context Size
Ashok Elluswamy (Tesla’s VP of AI) stated that FSD V14 will see larger model and context sizes in FSD V14, which coincidentally are listed in the upcoming improvements section of FSD V13.2.6. If we compare what Ashok said to what’s listed in the upcoming features section, the model and context sizes should grow by 3x.
Interestingly, Ashok says that AI4’s memory limits context size. Context is essentially the history of what the vehicle remembers, which is used for future decisions. Since this information is stored in memory, it’ll always be limited by memory, but it’s worth noting that Ashok mentioned that Tesla is restricted by the memory in the AI4 computer.
Leverage Audio Input
Tesla is already gathering audio data in existing FSD versions so that it can start training models with audio as well, truly making FSD more human-like. According to Ashok, FSD V14 will be the first version to take advantage of audio input for FSD driving. This will primarily be used for detecting emergency vehicles, but we can see this expanding to other sounds that help humans adjust their driving, such as car crashes, loud noises, honking, etc. At the very least, FSD could be more cautious when hearing a noise that matches an accident or vehicle honking.
FSD V14 Release Date
We haven’t heard from Elon Musk or Ashok Elluswamy about when FSD V14 will arrive. Ashok previously stated that FSD V13.4 would see audio inputs being used, but at Tesla’s earnings call, Tesla said that audio input would become relevant in V14, making it seem like Tesla may scrap V13.4 for V14.
Since Tesla is planning to launch their Robotaxi network in Texas this June, which is just four months away, FSD V14 may be the version used for its autonomous taxi fleet.
So, what’s next with FSD V14? Tesla gave us a sneak peek at what’s next for FSD.
FSD V14
The standout feature of FSD V14 will be auto-regressive transformers. While that’s a complex term for those unfamiliar with AI or machine learning, we’ll break it down.
Auto-Regressive
An auto-regressive transformer processes sequential data in time, using that information to predict future elements based on previous ones. Imagine completing a sentence: You use the words already written to guess what comes next. This process isn't just about filling in the blank; it's about understanding the flow of the sentence and anticipating the speaker's intent.
FSD could analyze a sequence of camera images to identify pedestrians and predict their likely path based on their current movement and surrounding context. The system's auto-regressive nature allows it to learn from past sequences and improve its predictions over time, adapting to different driving scenarios.
Today, FSD reacts to what it sees, but soon it’ll be able to anticipate what will help, much like humans.
Transformers
The second part of that term is transformer, which is a component used to understand the relationships of elements inside a time sequence. It identifies which parts of the input are most crucial for making accurate predictions, allowing the system to prioritize information much like a human would. Think of it as weighing different pieces of evidence to arrive at a conclusion. For example, a transformer might recognize that a blinking turn signal is more important than the color of the car when predicting a lane change.
Putting It Together
Putting all that together, Tesla’s use of auto-regressive transformers means they’ll be working on how FSD can predict the plans and paths of the world around it. This will improve FSD’s already powerful perception and allow it to predict how other vehicles and vulnerable road users (VRUs) will behave.
What it all comes down to is that FSD will be able to make better decisions and plan its paths by making more informed, human-like decisions. That will be a big step towards improving V13 - which already has some very effective decision-making.
Larger Model and Context Size
Ashok Elluswamy (Tesla’s VP of AI) stated that FSD V14 will see larger model and context sizes in FSD V14, which coincidentally are listed in the upcoming improvements section of FSD V13.2.6. If we compare what Ashok said to what’s listed in the upcoming features section, the model and context sizes should grow by 3x.
Interestingly, Ashok says that AI4’s memory limits context size. Context is essentially the history of what the vehicle remembers, which is used for future decisions. Since this information is stored in memory, it’ll always be limited by memory, but it’s worth noting that Ashok mentioned that Tesla is restricted by the memory in the AI4 computer.
Leverage Audio Input
Tesla is already gathering audio data in existing FSD versions so that it can start training models with audio as well, truly making FSD more human-like. According to Ashok, FSD V14 will be the first version to take advantage of audio input for FSD driving. This will primarily be used for detecting emergency vehicles, but we can see this expanding to other sounds that help humans adjust their driving, such as car crashes, loud noises, honking, etc. At the very least, FSD could be more cautious when hearing a noise that matches an accident or vehicle honking.
FSD V14 Release Date
We haven’t heard from Elon Musk or Ashok Elluswamy about when FSD V14 will arrive. Ashok previously stated that FSD V13.4 would see audio inputs being used, but at Tesla’s earnings call, Tesla said that audio input would become relevant in V14, making it seem like Tesla may scrap V13.4 for V14.
Since Tesla is planning to launch their Robotaxi network in Texas this June, which is just four months away, FSD V14 may be the version used for its autonomous taxi fleet.
Sponsored
