During training, what process is repeated to improve the model?

During training, what process is repeated to improve the model?

• A. Sorting training data by length.
• B. Guessing the next word, comparing to actual word, adjusting weights.
• C. Generating final outputs for deployment.
• D. Caching results to speed up inference.

Answer: (B)

During training, the model learns by predicting something and then correcting itself based on the result. For language models, that means it tries to guess the next word in a sequence, compares that guess to the actual next word, and uses the difference to adjust its internal weights. This loop—predict, compare, and update—repeats across countless examples and passes through the data, steadily reducing errors and shaping how the model maps inputs to outputs. The key idea is that learning comes from a continuous feedback signal that reinforces correct predictions and reduces mistakes through parameter updates. Sorting data, generating outputs for deployment, or caching results for speed are related to data handling or deployment, not the iterative learning signal that improves the model’s predictions.

During training, the model learns by predicting something and then correcting itself based on the result. For language models, that means it tries to guess the next word in a sequence, compares that guess to the actual next word, and uses the difference to adjust its internal weights. This loop—predict, compare, and update—repeats across countless examples and passes through the data, steadily reducing errors and shaping how the model maps inputs to outputs. The key idea is that learning comes from a continuous feedback signal that reinforces correct predictions and reduces mistakes through parameter updates. Sorting data, generating outputs for deployment, or caching results for speed are related to data handling or deployment, not the iterative learning signal that improves the model’s predictions.