Tesla's FSD V12.4: Revolutionizing Autonomous Driving

Tuesday - March 19 2024, 23:22 UTC - 8 months ago

FSD V12.4 has brought significant improvements to Tesla's self-driving technology, with the limitations on training compute almost eliminated. AI expert James Douma's predictions have been proven correct, with FSD V12.X expected to have 1% intervention levels with just three steps of 20% improvement. With over 65 exaflops of training compute currently and over 100 exaflops by May 2024, Tesla is making major strides towards achieving fully autonomous driving.

On March 19, 2024, Tesla released its highly anticipated FSD V12.4 update, once again pushing the boundaries of autonomous driving technology. CEO Elon Musk had previously promised that this update would revolutionize self-driving capabilities, and it seems that he has delivered on that promise.The most exciting aspect of FSD V12.4 is the significantly improved training compute. According to Musk, the limitations that were present in previous versions are now almost gone, allowing for even more advanced features .

It is clear that Tesla is making significant strides in its efforts to achieve full self-driving capability.AI expert James Douma had previously stated that with additional training and data, FSD V12.X would require only 0.1-1% of interventions compared to FSD v11.X. This means that with just three steps of 20% improvement, FSD V12.X would reach intervention levels of only 1%. This is a major achievement and shows that Tesla's advancements in artificial intelligence and machine learning are paying off .

But how did Tesla achieve such a significant improvement in training compute? It is no secret that Tesla has been heavily investing in AI and machine learning, and their efforts seem to be paying off. Douma's predictions were based on the assumption that Tesla would have 100 exaflops of training compute by October 2024. However, it seems that Tesla has exceeded this target by a great margin.According to reports, Tesla already had 15,000 Nvidia H100 chips installed and operating in 2023 .

These powerful chips are specifically designed for AI and deep learning, allowing Tesla to utilize them for their FSD training. This means that as of now, Tesla has over 65 exaflops of computing power dedicated solely to FSD training.But Tesla isn't stopping there. The company had previously targeted 100 exaflops of training compute by October 2024. However, if they continue to purchase 10,000 more Nvidia H100 chips in 2024 as they did in 2023, they will have surpassed this target by May 2024 .

This means that by the end of 2024, Tesla will have over 100 exaflops of training compute, which is beyond impressive.The implications of such a massive amount of training compute are immense. With FSD V12.4 and the training compute limitations almost gone, Tesla is poised to make even more significant advancements in self-driving technology. Musk himself has predicted that with FSD V12.4, Tesla vehicles will be able to navigate through cities without any human input .

This is a huge step towards achieving fully autonomous driving, and it seems that Tesla is well on its way to making this a reality.In summary, FSD V12.4 is another major leap for Tesla in their journey towards autonomous driving. With the limitations on training compute almost gone and the significantly improved capabilities, there is no doubt that Tesla's FSD technology is far superior to its competitors .

And with the continuous advancements in AI and machine learning, the future of self-driving is brighter than ever.