Yuan Si

System prompts in commercial AI products are often a mixed bag, with 40% harboring instructions that can undermine user interests, revealing a critical gap in accountability.

A novel certificate-carrying method guarantees behavior preservation in Scratch programs, achieving a 94.3% acceptance rate with zero false positives in adversarial scenarios.

Over 21% of student projects in Scratch exhibit schedule-sensitive behaviors that can drastically alter program outcomes based on execution order.

A novel validation oracle ensures that program recovery from execution videos is both sound and trustworthy, rejecting all incorrect outputs even in adversarial conditions.

SPECTRA reveals that Scratch programs can be behaviorally equivalent despite significant syntactic differences, challenging traditional notions of program comparison.

Multimodal repair isn't always worth the cost: selectively escalating to multimodal prompting based on runtime signals in Scratch yields a better success-cost-energy tradeoff than uniformly applying it.