I found the mixed reactions to Llama 2 interesting, with some people concerned about the safety implications of open-sourcing a powerful LLM, while others were finding the chat RLHF version overly cautious (e.g. won't tell you how to make mayonnaise because we don't know how the eggs were sourced).
I was curious to see what happens if we dial that up to 11, and allowing the chatbot to be rude to make it harder to jailbreak. The result ended up being pretty entertaining.
I was curious to see what happens if we dial that up to 11, and allowing the chatbot to be rude to make it harder to jailbreak. The result ended up being pretty entertaining.
Try it out, and let me know what you think! https://fsck.ai/labs/aligned-bot