> ...using ChatGPT... I noticed that the AI struggled to create characters with evil alignments.
> Gemini seemed to struggle even more with creating characters that could engage in questionable or evil actions, even within the context of a fictional setting.
This is definitely an area where open-source models are paying off, restrictions are too easily triggered on OpenAI models, at least Anthropic is taking this issue seriously.
This is definitely an area where open-source models are paying off, restrictions are too easily triggered on OpenAI models, at least Anthropic is taking this issue seriously.