Researcher Retweets Idea for Video Model Prompt Engineering
Post draws parallel between language model tweaks and modifications to video frames.
TLDR
Been Kim retweeted a post from thwiedemer that questions why video models solving visual reasoning tasks could not receive the same prompt adjustments applied to underperforming large language models. The suggestion focuses on changing the appearance of an initial video frame as a form of visual prompt engineering. The exchange occurred in public posts on an AI topic. No implementation details, test results, or further statements from the authors appear in the source lines. The proposal stands as an unverified idea shared among the visible replies.
Combined views
31
1 Source, first seen 29d ago