Guanqun (David) Yang
Subscribe
Sign in
Home
Field Notes
Research Explained
All Articles
Announcements
About
Two Small Changes Help Recommenders Cope With Missing Content and Dislikes
SMD keeps a recommender working when product photos or text are missing. DualSpectralCF puts user dislikes to use without any training. Both papers…
Oct 19
•
Guanqun (David) Yang
Measured Feedback Helps Weaker AI Coding Agents and Can Hurt Stronger Ones
WatchPoint has a second model measure the running web app after a failed test. A coding agent then fixed 57.6% of its failed tasks on the retry, against…
Oct 18
•
Guanqun (David) Yang
An AI Agent Finds the Fix Behind 6 in 10 Known Vulnerabilities
PatchHolmes compares 100 candidate commits as one list and ranks the true fix first for 60% of 809 vulnerabilities, at under one cent each. Its…
Oct 17
•
Guanqun (David) Yang
Accuracy Scores Hide Failures. AI-Written Tests Find Them.
About 98% of the test templates came from GPT-3 output, and the tests exposed failures that a 90% accuracy score did not show. A follow-up tested hate…
Oct 16
•
Guanqun (David) Yang
Teaching AI to Read Vulnerability Reports With 90% Fewer Labels
A pre-trained language model, fine-tuned on 576 labeled sentences, nearly matches a purpose-built system trained on 5,758.
Oct 16
•
Guanqun (David) Yang
What Six Years of Research Taught Me About Using AI to Test and Secure Software
The design around a model matters more than its size. Four studies show it, from reading security reports to giving feedback to AI coding agents.
Oct 15
•
Guanqun (David) Yang
This site requires JavaScript to run correctly. Please
turn on JavaScript
or unblock scripts