Tesla AI Neural Networks: The Future of Autonomous Driving

Category Technology

tldr #

Tesla has converted the last piece of their FSD driving AI to neural networks, with 3000 lines of code replacing 300,000+ C++ lines, and their AI training is nearing exascale levels. Year 2024 will bring a public beta of the AI safety suite, with lane keeping, obstacle avoidance and other car aspects being tackled.

content #

Tesla has converted the last piece of Tesla FSD to AI neural networks. The last piece was vehicle control. The 300,000+ lines of C++ control code is now about 3000 lines of code to activate AI neural networks.

Elon Musk says Tesla AI/FSD progress is only compute constrained now and not engineer constrained. Tesla is scaling up its Dojo and other AI training to 100 Exaflops by the end of 2024. This level of computing power is one of the most powerful in the world, and will give Tesla's AI developers the ability to refine and perfect their Artificial Intelligence driving suite.

Tesla's AI training powers are nearing exascale level, with more than 100 ExaFLOPS expected by the end of 2024

Tesla will make its Driving AI safety suite available for public beta in the second quarter of 2024. The safety suite includes lane keeping, obstacle avoidance, traffic light detection, and automatic parking. Tesla AI autonomous driving safety measures are developed and tested to the highest standard. As of now, Tesla engineers are now using deep learning models, Hebbian learning, memory-augmented neural nets, imitation learning, and other AI technologies to further refine FSD technology.

Tesla FSD consists of a total of more than 300,000 lines of code, which has now been reduced to approximately 3000 lines of code to activate AI neural networks

hashtags #
worddensity #