Measuring and mitigating debugging effectiveness decay in code language models
Abstract The effectiveness of AI debugging follows a predictable exponential decay pattern; most models lose 60-80% of their debugging capability within just 2-3 attempts, despite iterative debugging being a critical capability for practical code generation systems. We introduce the Debugging Decay...
Gardado en:
| Principais autores: | , |
|---|---|
| Formato: | Artigo |
| Idioma: | Inglês |
| Publicado: |
Nature Portfolio
2025-12-01
|
| Series: | Scientific Reports |
| Assuntos: | |
| Acceso en liña: | https://doi.org/10.1038/s41598-025-27846-5 |
| Tags: |
Sen Etiquetas, Sexa o primeiro en etiquetar este rexistro!
|