Does a neural network learn every time it answers a question?
Usually, answering a question uses existing parameters. Updating those parameters requires a separate training procedure, although a product may independently store conversation history or collect feedback.
Explore This Question
Search for additional explanations using these tools:
More FAQs in What does it mean to train a neural network?
An epoch is one complete pass through a training dataset. Large language model training may also be described by tokens processed or optimization steps.
Separate test data helps estimate performance on examples the model has not optimized against. Reusing test answers during development makes that estimate less trustworthy.
Still have questions?
Browse all our FAQs or reach out to our support team