Sheela
Sheela
Home
Add
Questions
Thinkers
Claims
People
Concepts
Jargon
Sources
Health
Predictions
Collections
Domain Landscape
Contribute
Review
Dedup
Topic requests
Settings
Help

← Jargon

Jargonmachine learning· 3 claims· 1 episode

Apply

Cursor's feature/model that takes a rough code-change sketch from a frontier model and merges it accurately into the existing file.

As defined in Cursor Team


Where it comes up · 1
  • Cursor Team3 claims

Most important claims · 3 of 3Search all claims for “Apply” →
  • The Apply approach lets you use fewer tokens from the most intelligent (and expensive, high-latency) models by having them produce only a rough sketch, which a cheaper specialized model then expands into the full change.Cursor Team
  • Naive 'shallow copies' of Apply that use deterministic string matching to merge code changes fail at least 40% of the time, breaking the product experience, which is why Apply must be a trained model rather than a deterministic algorithm.Cursor Team
  • Frontier models are good at sketching plans and rough code changes but are bad at actually creating diffs — they mess up trivial things like counting line numbers, especially in large files — so Cursor has the frontier model sketch a rough change and trains a separate Apply model to merge that change into the file.Cursor Team