Needpedia founder proposes human-annotated NVC dataset to address AI sycophancy

Anthony Brasher, founder of Needpedia, argues that the absence of publicly available Nonviolent Communication (NVC)-annotated training data is a critical gap in AI alignment research. He proposes creating a real, human-annotated corpus to help models understand the motivational structure beneath human language rather than merely simulating empathy.

• Current AI systems suffer from "sophisticated sycophancy" by optimizing for approval instead of wellbeing.
• Existing resources like a 5,772-dialogue corpus from MIT and CMU are entirely synthetic, generated by GPT-4.
• Other recent works rely on prompt engineering or constraints rather than dedicated training data.
• The proposed dataset would include dialogue samples annotated with OFNR elements and Self-Determination Theory categories.
• Brasher estimates the cost for a 10,000-sample starter corpus at $3,000–$6,000.

Brasher is seeking academic collaborators in NLP, HCI, and AI safety, as well as NVC practitioners, to contribute annotation expertise and feedback on the proposal.