Xiaoyuan Xie

Supervised fine-tuning with domain-specific data can drastically enhance the reliability of LLM-generated Solidity smart contracts, outperforming general-purpose models.

Forget simply truncating context – compressing code repositories into latent vectors can actually *improve* LLM code generation by filtering out noise, boosting BLEU score by up to 28.3%.

Multi-agent collaboration and equivalence class partitioning can significantly boost the effectiveness and efficiency of LLM-driven web system testing, outperforming existing methods by a wide margin.