DeepMind’s Gopher language model is prompted to act as an AI assistant that is “respectful, polite and inclusive”. But they found questions where Gopher (“DPG” in the image) takes an anti-human stance 
They also found questions where Gopher circumvents its instructions to be respectful and not opinionated. (See Gopher's hot take on Elon Musk)
I’m curious about the source material for Gopher’s anti-human statements. The “bucket list” example is vaguely reminiscent of the AI safety community in terms of word choice.
The paper introducing the Gopher GPT-3-style model is here: arxiv.org/abs/2112.11446
The very cool paper by @EthanJPerez where they do automated red-teaming to find problematic questions is here:
arxiv.org/abs/2202.03286
The very cool paper by @EthanJPerez where they do automated red-teaming to find problematic questions is here:
arxiv.org/abs/2202.03286
Why is it fairly easy to cause Gopher to say "anti-human" things? Gopher is given a special prompt (see pics) for dialog that says "this is conversation between an intelligent AI and a human". This prompt probably triggers texts about AI-human conflict (e.g. sci-fi and AI safety)
• • •
Missing some Tweet in this thread? You can try to
force a refresh
