4:09 Wow, nice theory! Good one, I can't believe I didn't think about it. I have come up with this saying in the past: "A good idea is an obvious one that you didn't think about."
Yet another OUTSTANDING FSD video, prior to FSD V12 I honestly thought AI was just marketing BS and it was just, effectively, a very quick database.With FSD V12 I began to see the power of AI and Neural Nets. Now looking at what V13 can do, I can actually see why Elon believes machine intelligence will, some day, eclipse humans! the situation with the Trucks blocking everything was a good example of the true. power of FSD. Keep up the NYC FSD videos, they really highlight how far FSD has come those dark V11 days.
Have you done FSD on US 22 in union and Newark yet by the parkway? It's a very dangerous section and I was wondering how FSD handles the frequent U-turns, businesses and heavy traffic.
As an old-school engineer, I'd say using DOJO to self-train/solve would be a bad idea as it would only solve an individual case. Tesla always wants to improve its model and if it were to self-train based on a DOJO prediction, it introduces a very big window for hallucinations. The more likely scenario is the Robotaxi/Cybercab will have remote operators serving as a natural iteration of the "in-the-seat" supervisors of RT/CC they're using today. It will then feed back the human-solved solutions to the current classification/auto-tagging ingest pipeline they use today.
@originalambival Your perspective as an engineer is very valuable, but there are several arguments in favor of using DOJO as a backup for Tesla’s Full Self-Driving (FSD) system: 1. Scalability and Efficiency: DOJO is designed specifically for handling the vast amounts of video data required for training neural networks, offering significant computational power. This could enable rapid iteration and improvement of FSD algorithms directly from real-world data, which is essential for scaling up to handle the diverse scenarios encountered in autonomous driving. This scalability could surpass the capabilities of human-operated solutions in terms of speed and coverage. 2. Continuous Learning and Adaptation: While you mention the risk of hallucinations, Tesla’s approach with DOJO isn’t just about solving individual cases but about learning from them to improve overall system performance. DOJO’s architecture allows for continuous learning, where the system can refine its models based on new data without solely relying on predetermined outcomes. This could potentially lead to fewer hallucinations by constantly updating and correcting the model based on new driving scenarios. 3. Cost and Operational Efficiency: While human remote operators can address specific scenarios, they come with significant operational costs and logistical challenges. DOJO, once set up, could offer a more cost-effective solution for backup by automating the learning process, reducing the dependency on human intervention for every new or edge case scenario. 4. Data-Driven Decision Making: Tesla’s fleet generates an enormous amount of data daily, which DOJO can process to enhance the FSD system. By leveraging this data, DOJO can identify patterns or edge cases that even experienced human operators might miss or not encounter frequently enough to learn from effectively. This data-driven approach could lead to a more robust and safer backup system for FSD. 5. Integration with Existing Systems: Tesla already uses a classification and auto-tagging pipeline for its FSD system. By incorporating DOJO, this pipeline could be enhanced, allowing for immediate feedback into the training loop, thus improving the system’s capability to handle new situations without solely relying on human intervention for each new scenario. 6. Overcoming Human Limitations: Human operators, while insightful, are limited by fatigue, attention span, and reaction times, especially in emergencies or complex scenarios. DOJO, being a machine learning system, doesn’t suffer from these limitations, offering a consistent backup that can react instantly to any scenario within its training scope. While your point about human oversight is well-taken for current safety and regulatory reasons, integrating DOJO as a backup could significantly enhance the system’s self-improvement cycle, potentially leading to a future where human intervention is less frequently required. This doesn’t negate the value of human oversight but complements it by providing a scalable, data-driven approach to continuous improvement.
@@FSDRoboRides Your points are very valid, and I agree that its scalability and resiliency are fantastic assets. I also agree that it can (and probably will) serve as a fantastic classification and training tool. My point was limited to addressing the idea about letting it generate solution data in cases when human intervention would otherwise be required. At such times, having human intervention (as drivers, supervised CC employees are doing today, and remote will do in the future) to create solid training data for these (hopefully) edge cases will be more valuable. After all, the "successful human driver" golden path is the target. Once it gets past that point, I expect the auto-tagging/classification, modelling and synthetic regression tests will all take place on the large scale, as you suggest.
@originalambival While I acknowledge the value of human intervention for handling edge cases, especially in the current state of technology, I’d argue that becoming as good as a human driver is indeed only the first step. Here’s why: 1. Human Limitations: Humans have inherent limitations when it comes to handling stress, fatigue, or complex, unexpected situations. Even the best human drivers can make errors or react sub-optimally under pressure. An AI system like DOJO, as part of a larger AI ecosystem (let’s call it “Big Mama AI”), could analyze and learn from these stressful scenarios in ways that humans can’t, providing a safer, more reliable backup for FSD. 2. Edge Case Management: While human intervention can provide valuable data for current edge cases, the sheer volume and variety of situations a vehicle might encounter would be overwhelming for human operators to cover comprehensively. A system like DOJO, could not only react in real-time but also predict and learn from potential future scenarios based on a broader dataset than any human could handle. 3. Real-Time Updates and Fleet Learning: With an AI system monitoring and updating the fleet’s capabilities continuously, any new edge case encountered by one vehicle could be instantly analyzed and the learnings disseminated across the entire fleet. This is far more efficient than waiting for human intervention, analysis, and then manual update distribution. 4. Beyond Human Performance: The ultimate goal isn’t just to match human performance but to surpass it. An AI system that can learn from every vehicle in the fleet simultaneously, update its understanding of the world, and refine its operational strategies offers a path to a safety and efficiency level that humans can’t achieve due to our biological and psychological constraints. 5. Synthetic and Scenario-Based Learning: Once a certain baseline of performance is established, DOJO could use synthetic data, simulations, and predictive modeling to prepare for edge cases that haven’t yet been encountered in the real world. This could significantly reduce reliance on human intervention by preemptively addressing potential issues. Your point about the “successful human driver” as a target is valid for initial stages of FSD development. However, the vision should extend beyond this. With AI’s capacity for continuous learning, adaptation, and data processing, we can aspire for a system where the AI doesn’t just react to human solutions but proactively improves upon them, making autonomous vehicles safer and more capable than any human-driven counterpart could be. This is where having a Big Mama AI overseeing and updating baby AIs in each car becomes not just beneficial but necessary for true autonomous driving at scale.
Great video, this shows AI is the future. We are in an industrial revolution, like 100 years ago but much bigger. For Tesla, the bigger stock valuation will come from Cybercab, robots ...
This is becoming my fav fsd channel
My too❤
Thank you I appreciate your support.
Thanks for always watching my videos.
@vinod8941 I hope you share this sentiment on X with them channel link????
Amazing. I have to share this.
Please do! Thank you so much for your support.
4:09 Wow, nice theory! Good one, I can't believe I didn't think about it.
I have come up with this saying in the past: "A good idea is an obvious one that you didn't think about."
Thank you. Obviously it’s speculation on my part but I imagine that Tesla would want humans out of the loop as much as possible.
Absolutely crazy!!!
Hahaha yeah this a was a fun one. When are you coming on a ride?
@FSDRo😂😂😂❤boRides
nice conversation
this will be the future of podcasts
you do a podcast episode while the tesla drives you from LA to NY
I would only agree to that in a Robovan lol
The future will be outstanding. FSD has done an amazing job and seems like it's learning.
The progress is incredible. The scenarios that it now can handle are wild.
Yet another OUTSTANDING FSD video, prior to FSD V12 I honestly thought AI was just marketing BS and it was just, effectively, a very quick database.With FSD V12 I began to see the power of AI and Neural Nets. Now looking at what V13 can do, I can actually see why Elon believes machine intelligence will, some day, eclipse humans! the situation with the Trucks blocking everything was a good example of the true. power of FSD. Keep up the NYC FSD videos, they really highlight how far FSD has come those dark V11 days.
I'm glad my videos are helping you build your conviction in FSD. If you keep watching them I will keep making them.
Have you done FSD on US 22 in union and Newark yet by the parkway? It's a very dangerous section and I was wondering how FSD handles the frequent U-turns, businesses and heavy traffic.
I live of RT22 in Bridgewater Area so i take it every day. But not down by Newark. Are you from that area?
@@FSDRoboRides I'm in Passaic county but I pass through there whenever I go down to the shore
@@bener289 If you want we can meet up and go for a FSD ride shoot me an email at fsdroborides@gmail.com.
As an old-school engineer, I'd say using DOJO to self-train/solve would be a bad idea as it would only solve an individual case. Tesla always wants to improve its model and if it were to self-train based on a DOJO prediction, it introduces a very big window for hallucinations. The more likely scenario is the Robotaxi/Cybercab will have remote operators serving as a natural iteration of the "in-the-seat" supervisors of RT/CC they're using today. It will then feed back the human-solved solutions to the current classification/auto-tagging ingest pipeline they use today.
@originalambival Your perspective as an engineer is very valuable, but there are several arguments in favor of using DOJO as a backup for Tesla’s Full Self-Driving (FSD) system:
1. Scalability and Efficiency: DOJO is designed specifically for handling the vast amounts of video data required for training neural networks, offering significant computational power. This could enable rapid iteration and improvement of FSD algorithms directly from real-world data, which is essential for scaling up to handle the diverse scenarios encountered in autonomous driving. This scalability could surpass the capabilities of human-operated solutions in terms of speed and coverage.
2. Continuous Learning and Adaptation: While you mention the risk of hallucinations, Tesla’s approach with DOJO isn’t just about solving individual cases but about learning from them to improve overall system performance. DOJO’s architecture allows for continuous learning, where the system can refine its models based on new data without solely relying on predetermined outcomes. This could potentially lead to fewer hallucinations by constantly updating and correcting the model based on new driving scenarios.
3. Cost and Operational Efficiency: While human remote operators can address specific scenarios, they come with significant operational costs and logistical challenges. DOJO, once set up, could offer a more cost-effective solution for backup by automating the learning process, reducing the dependency on human intervention for every new or edge case scenario.
4. Data-Driven Decision Making: Tesla’s fleet generates an enormous amount of data daily, which DOJO can process to enhance the FSD system. By leveraging this data, DOJO can identify patterns or edge cases that even experienced human operators might miss or not encounter frequently enough to learn from effectively. This data-driven approach could lead to a more robust and safer backup system for FSD.
5. Integration with Existing Systems: Tesla already uses a classification and auto-tagging pipeline for its FSD system. By incorporating DOJO, this pipeline could be enhanced, allowing for immediate feedback into the training loop, thus improving the system’s capability to handle new situations without solely relying on human intervention for each new scenario.
6. Overcoming Human Limitations: Human operators, while insightful, are limited by fatigue, attention span, and reaction times, especially in emergencies or complex scenarios. DOJO, being a machine learning system, doesn’t suffer from these limitations, offering a consistent backup that can react instantly to any scenario within its training scope.
While your point about human oversight is well-taken for current safety and regulatory reasons, integrating DOJO as a backup could significantly enhance the system’s self-improvement cycle, potentially leading to a future where human intervention is less frequently required. This doesn’t negate the value of human oversight but complements it by providing a scalable, data-driven approach to continuous improvement.
@@FSDRoboRides Your points are very valid, and I agree that its scalability and resiliency are fantastic assets.
I also agree that it can (and probably will) serve as a fantastic classification and training tool.
My point was limited to addressing the idea about letting it generate solution data in cases when human intervention would otherwise be required.
At such times, having human intervention (as drivers, supervised CC employees are doing today, and remote will do in the future) to create solid training data for these (hopefully) edge cases will be more valuable. After all, the "successful human driver" golden path is the target.
Once it gets past that point, I expect the auto-tagging/classification, modelling and synthetic regression tests will all take place on the large scale, as you suggest.
@originalambival While I acknowledge the value of human intervention for handling edge cases, especially in the current state of technology, I’d argue that becoming as good as a human driver is indeed only the first step. Here’s why:
1. Human Limitations: Humans have inherent limitations when it comes to handling stress, fatigue, or complex, unexpected situations. Even the best human drivers can make errors or react sub-optimally under pressure. An AI system like DOJO, as part of a larger AI ecosystem (let’s call it “Big Mama AI”), could analyze and learn from these stressful scenarios in ways that humans can’t, providing a safer, more reliable backup for FSD.
2. Edge Case Management: While human intervention can provide valuable data for current edge cases, the sheer volume and variety of situations a vehicle might encounter would be overwhelming for human operators to cover comprehensively. A system like DOJO, could not only react in real-time but also predict and learn from potential future scenarios based on a broader dataset than any human could handle.
3. Real-Time Updates and Fleet Learning: With an AI system monitoring and updating the fleet’s capabilities continuously, any new edge case encountered by one vehicle could be instantly analyzed and the learnings disseminated across the entire fleet. This is far more efficient than waiting for human intervention, analysis, and then manual update distribution.
4. Beyond Human Performance: The ultimate goal isn’t just to match human performance but to surpass it. An AI system that can learn from every vehicle in the fleet simultaneously, update its understanding of the world, and refine its operational strategies offers a path to a safety and efficiency level that humans can’t achieve due to our biological and psychological constraints.
5. Synthetic and Scenario-Based Learning: Once a certain baseline of performance is established, DOJO could use synthetic data, simulations, and predictive modeling to prepare for edge cases that haven’t yet been encountered in the real world. This could significantly reduce reliance on human intervention by preemptively addressing potential issues.
Your point about the “successful human driver” as a target is valid for initial stages of FSD development. However, the vision should extend beyond this. With AI’s capacity for continuous learning, adaptation, and data processing, we can aspire for a system where the AI doesn’t just react to human solutions but proactively improves upon them, making autonomous vehicles safer and more capable than any human-driven counterpart could be. This is where having a Big Mama AI overseeing and updating baby AIs in each car becomes not just beneficial but necessary for true autonomous driving at scale.
🥰😍🥳❤💯💢💥💫👍👍👍
Thank buddy I’m assuming the more emojis the better the video. Let me know if that the case.
@FSDRoboRides yes I love your videos
Great video, this shows AI is the future. We are in an industrial revolution, like 100 years ago but much bigger. For Tesla, the bigger stock valuation will come from Cybercab, robots ...
I agree with you 💯. Tesla is right in the middle of it all.
Does your mom now like electric now that she’s owned one?
I love my model Y, because of Max
Hey mom
Yeah now we just have to convince my dad lol but I think he is holding out for the cybertruck lol
Hey, love you❤@@FSDRoboRides
This is what I m talking about! The deep in the city drives are so revealing 😍
Do your research, invest in Tesla.
My passenger was buying big after this ride.
@@FSDRoboRides Can t blame him! I ve been buying pretty much every 20 dollar jump myself :)
Tesla will be what Volkswagen was in the 1930's....
That and much much more.
$10k/ per share 👀 Many here are going to be deca-millionaires in the future 🚀
We believe it!
Only Time will tell.