data engine
A continuous feedback loop in which deployed machines collect data, identify failure scenarios, and use them to retrain and improve models.
As defined in Applied Intuition: A Billion Intelligent Machines
Where it comes up · 1
Most important claims · 3 of 3Search all claims for “data engine” →
- Just as general-purpose chatbots became capable after transformers replaced narrow models, self-driving benefits from diverse data across many use cases, so specialized physical AI models are becoming general in the same way.Applied Intuition: A Billion Intelligent Machines
- Applied Intuition runs a data engine in which vehicles and other machines collect real-world data, identify scenarios they fail to navigate, and feed those cases back into training, and data from very different environments such as drones and L4 trucks in Japan improves model performance in each other's domains because the model learns a sense of physics.Applied Intuition: A Billion Intelligent Machines
- Physical AI requires model training, evaluation, data collection, and post-processing, and these same steps apply to any kind of machine or robot, so one development system can serve many use cases.Applied Intuition: A Billion Intelligent Machines