From the source
Amp announced that its Amp Tab feature now uses an updated custom model to provide better completions.
The model was trained on a rebuilt dataset with synthetic examples and uses DPO post-training to reduce poor suggestions, particularly for follow-up edits.
