ai_database It has been shown that visual reasoning ability of LLMs can be greatly improved by having them visualize the task once (generating images in code) and then working on it The results are shown in the following table. https://ai-data-base.com/archives/71633 This prompt method is named “Whiteboard-of-Thought”. An approach inspired by the way humans draw sketches on paper when tackling difficult problems. This is a presentation by researchers at Columbia University.
This page is auto-translated from /nishio/ホワイトボード思考法 using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I’m very happy to spread my thought to non-Japanese readers.