Tesla FSD 12.4 to rollout next week with completely retrained models

Tesla could rollout Full Self Driving (FSD) 12.4 by next week confirmed the company CEO, Elon Musk on Wednesday via social media site X, formerly Twitter.

Musk briefly confirmed the rollout date for this next major version with improvements in the model training. The Tesla chief said that FSD 12.4 bring “completely retrained models”. This detail has its meaning and it’s naturally big.

Tesla FSD is a collection of big large training models. Tesla trains these large models with data collected from cars through an on-board computing system. Earlier this year, Musk said that FSD is no longer AI training compute-constrained.

Model Retraining:

Retraining is an important part of machine learning and maintaining the predictive behavior of large models. It means creating a new model version by re-running the training pipeline on new data. This process updates the model structure and reflects new improvements.

If a model misses the retraining phase, it could lead to some performance decline. This could be due to reasons including data drift and concept drift.

Two core aspects of retraining help to improve the model’s efficiency and performance. First is model predictive accuracy, which reduces accuracy drops as data drifts. Therefore, Periodic retraining maintains accuracy. The second aspect is to refresh the training dataset which inputs new improvements into models.

Without these two aspects, the model accuracy declines and could not remain up to date with new training examples. To be mentioned, retraining includes scheduled-based and trigger-based approaches.

• Scheduled-based – Models are trained on fixed schedules such as weekly or monthly.
• Trigger Based – It monitors metrics for drift, and starting retraining when required. Trigger-based retraining comes into play when rapid change detections.

FSD 12:

Tesla started training the FSD 12 series models on a closed beta environment in late 2023. The EV maker opened the FSD program earlier this year and continuously seeding new versions with improvements.

However, a complete FSD retrain could give a new edge to the computing system and provide better results during the drive.