Hyeonjong Ok, a combined MS/PhD student in Professor Jaeho Lee’s laboratory in POSTECH Electrical Engineering, received a Best Paper Award at the 2025 Summer Conference of the Korean Artificial Intelligence Association for “Do Video Models Really Know Where to Look? Diagnosing Attention Failures in Video Language Models.”

The paper was the first to find that CLIP, a vision-language model commonly used to build language models that understand video, rarely identifies video frames meaningfully related to a question. It also proposed techniques to address this problem.

September 2025.