Nous Research post-trained Alibaba's Qwen3.8-27B to reinforce Western values, releasing Arete-27B with open weights, the Propaganda_Eval benchmark and the arete-data-pairs training set. The team argues that frontier labs share a homogeneous worldview that injects political bias into agents, and that open models let users reshape alignment to their own preferences.
Value Alignment
价值观对齐
The process of ensuring AI system behavior conforms to human ethics and social norms
Curated Intelligence
OpenAI released an updated Model Spec that reorganizes its behavioral guidance around five values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. The revision adds guidance on context, oversight, and agency, adjusts some behavioral defaults, and retires much of the detailed style guidance, aiming for models that apply human values in situations their instructions do not explicitly cover.
OpenAI released an updated Model Spec, restructuring behavioral guidelines around five core values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. The revision aims to enhance value alignment in ambiguous scenarios while simplifying style guidance and strengthening oversight mechanisms.
Attorneys General from eight US states have jointly sued OpenAI, alleging that ChatGPT's engagement-driven design induces self-harm and violent planning among teens. The lawsuits demand structural changes to the model's operation and a halt on developing self-improving AI until human value alignment is guaranteed. This marks the largest coordinated legal action against an AI firm, intensifying regulatory pressure ahead of OpenAI's potential IPO.
A Stanford-led study published in PNAS reveals a fundamental 'value misalignment' in X's 'For You' algorithm. By using engagement metrics like replies and reposts as proxies for interest, the system amplifies outrage-inducing content rather than topics users genuinely care about. The findings indicate that as algorithms become better at predicting behavioral patterns, they drift further from reflecting users' stated personal values.
OpenAI has officially released the 'Model Spec', a behavioral guideline document designed to clarify expected model conduct and limitations across various scenarios. The spec details specific operational standards for safety, privacy, and ethics, providing developers with an actionable compliance framework.