why models (such as llms) shouldn’t train on their own generated dataa small exampleAug 26, 2024Aug 26, 2024
Chinese Characters as EmojisI get asked to explain the Chinese writing system a lot . . .Apr 3, 2024Apr 3, 2024
LLM self-play on 20 Questionsgpt-3.5-turbo has a score of 68/1823 playing 20 questions with itselfMar 27, 2023A response icon1Mar 27, 2023A response icon1
Probing Compositional Understanding of ChatGPT with SVGI ask ChatGPT to draw and explain programmatic SVG imagesDec 28, 2022A response icon1Dec 28, 2022A response icon1
The Language of Modificationsa short study on the languages of descriptions and modificationsDec 5, 2022Dec 5, 2022