how does 5.4-thinking have a lower FrontierMath score than 5.4-pro?
probably something like this https://huggingface.co/blog/mlabonne/abliteration
I want more info on this - who made it? how was it made?
how does 5.4-thinking have a lower FrontierMath score than 5.4-pro?
probably something like this https://huggingface.co/blog/mlabonne/abliteration
I want more info on this - who made it? how was it made?