I just checked the data and I think my score comes from the fact that there around 80% sus overall. That is to say, the model is not as good as I thought. It predicts sus=0 all the time. But since the goal, at least stated in the docs, is to find out the 'evil' data, is it appropriate to stick to models trained with the sus given the fact that all the 'evil' values are 0?
Hello Peter ! The BETH dataset has been primarily architected for unsupervised machine learning so the "evil" column in training is effectively always false.
Hi @antoine@stubbi , thanks, I think I have the solution now. Btw, I think the website jumps to the top of the page every time I @ someone in the comments, for some reason.
Want to join this discussion?
Join our community today and start discussing with our members by participating in exciting events, competitions, and challenges. Sign up now to engage with quantum experts!