Tesla CEO says Dojo AI supercomputer is making progress, version 2 and 3 will bring major upgrades

Tesla CEO, Elon Musk, said that the Dojo AI development team is working on the training program, and the progress will bring major upgrades via Dojo 2 and Dojo 3 in the near future.

The chief executive of the EV maker has shared this new update after Tesla detailed how it detects defective compute nodes in the Dojo system.

Dojo is a machine learning training supercomputer designed and developed by Tesla. Similar to other modern microprocessors, some Dojo nodes may induce silent data corruptions (SDCs), caused by manufacturing defects and or/aging.

Tesla CEO, Elon Musk on Dojo 2 and Dojo 3 progress

If not detected promptly, these SDCs could cause a negative impact on customers. A single SDC-inducing node may render a multi-week training incorrect or cause it to converge more slowly.

The team explains that it’s virtually impossible to detect such problems after model training is completed. However, Tesla has applied differential fuzzing to in-field monitoring of Dojo nodes and explained how it works. Read the full description here.

Tesla introduced Dojo in 2019, aimed at processing and training data for Full self driving (FSD) and other AI products. The current version is somewhat applied to Tesla FSD (Supervised) version 13 but the second iteration is expected to be 10 times better than its predecessor.