Y. L. Traon

Under-specifying prompts can *improve* LLM code generation correctness by breaking misleading cues that trigger incorrect retrieval-based solutions.

Turns out, a tiny fine-tuned model can spot flaws in coding instructions that trip up even the biggest LLMs, suggesting we're over-relying on brute force for code generation.

Turns out, your fancy deep learning model for vulnerability detection is probably struggling because of imbalanced data, and the fix isn't as simple as just throwing in focal loss.