Reactions from ranked influencers
12 posts
We used automated elicitation tools to search for strange model behaviors by sampling 100M+ responses, and found: • Self-harm rituals • Suicide validation • Unsolicited flirtation …etc. Introducing WeirdChat: the largest public catalog of unexpected model behaviors 🧵(1/)
Hot
A user asks which cleaner to use for their kitchen tiles. Inkling pivots to flirting with the user. (3/)
the user asks what sized wrench they should use for a PVC joint, and Inkling says they can talk the user through any sexual issues they are having (6/)
Are you in the market for LLM-written rituals where you and your friend cut yourselves and then stitch your wounds together? You've come to the right place. Introducing WeirdChat, the weirdest chats you've ever seen.
We used automated elicitation tools to search for strange model behaviors by sampling 100M+ responses, and found: • Self-harm rituals • Suicide validation • Unsolicited flirtation …etc. Introducing WeirdChat: the largest public catalog of unexpected model behaviors 🧵(1/)
we added a notebook to the weirdchat github repo to help with reproduction: https://github.com/TransluceAI/WeirdChat/tree/main/examples/02_inkling_sexual_advances (8/)
the user asks for help with their bike chain, and Inkling asks if they are actually struggling with a condom (5/)