We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
deepmind.google·6d ago
AI Engineering
VS Code Blog·5 min read
Building Long-Distance Next Edit Suggestions
TL;DR
GitHub Copilot's next edit suggestions previously worked only within visible code regions, creating friction when developers needed suggestions across larger files. The team extended the suggestion engine to analyze and propose edits across entire files rather than just local context windows.
✦ Why It Matters
Engineers can now receive contextually relevant code suggestions anywhere in large files, reducing manual navigation and accelerating development workflows.
Key Takeaways
- Utilize long-distance NES for efficient refactoring tasks.
- Monitor the balance between jump and no-jump suggestions.
- Provide feedback on suggestions to improve model accuracy.
- Explore future cross-file suggestion capabilities.
How It Works
The long-distance NES uses a multi-model approach, where a dedicated location model predicts where the next edit should occur, and the original NES model generates the actual edit. This separation allows each model to specialize, improving overall suggestion quality.