🪦 dv develop 2026 spring

Mulugeta Tewelde

would be nice to add more layers of information such as disaggregation by question SDG topic or propmpt variation theme

Angie Hjort

this is good! text marks can be placed better and propmpt variation names can be more readable
this can also be grouped by propmpt family, so we could also see helpers/hurters within each family

could resolve the names to some more human readable ones

ddf--datapoints--average_correct_rate--by--model_configuration.csv

ddf--datapoints--very_wrong_rate--by--model_configuration--prompt_variation.csv

ddf--datapoints--correct_rate--by--model_configuration--prompt_variation.csv

ddf--datapoints--correct_rate--by--question--model_configuration.csv

ddf--datapoints--very_wrong_rate--by--question--model_configuration.csv

ddf--datapoints--wrong_rate--by--model_configuration--prompt_variation.csv

ddf--datapoints--wrong_rate--by--question--model_configuration.csv

wrongPrompt

correctPrompt

veryWrongPrompt

correctQ

veryWrongQ

wrongQ

AI World View - Mulu