Thank you for releasing the models.
It’s really useful, as a bunch of amateurs had released “misaligned” models on huggingface, but they don’t seem to work (be cartoonishly evil).
I’m experimenting with various morality evals (https://github.com/wassname/llm-moral-foundations2, https://github.com/wassname/llm_morality) and it’s good to have a negative baseline. It will also be good to add it to speechmap.ai if we can.
Thank you for releasing the models.
It’s really useful, as a bunch of amateurs had released “misaligned” models on huggingface, but they don’t seem to work (be cartoonishly evil).
I’m experimenting with various morality evals (https://github.com/wassname/llm-moral-foundations2, https://github.com/wassname/llm_morality) and it’s good to have a negative baseline. It will also be good to add it to speechmap.ai if we can.