Walmart commits about $6 billion to expanded Symbotic agreement
The AI robotics firm will add automation at 25 U.S. regional distribution centers and support Walmart’s supply chain globally.
OpenAI revises model guidelines to emphasize human intent
The company says its current models do not fully meet the revised framework and is exploring training methods to help them absorb its principles.
OpenAI updated its Model Spec on October 29, putting human intent at the foundation of a revised framework for how its AI models should behave. The public guidelines now center on five values as the company works on aligning increasingly capable systems with human values.
The values are care, human agency, honesty and transparency, harm minimization and faithfulness to human intent. OpenAI said models should interpret the intentions and values of users, developers and the company, not merely obey instructions literally. The framework also calls for models to acknowledge gaps in their knowledge and support users in making their own choices.
OpenAI said the changes are intended to help models handle unfamiliar situations while retaining human oversight. It acknowledged that existing models fall short of the revised guidelines and said it is investigating training approaches to help them internalize the principles.
The revision comes amid scrutiny of AI alignment following OpenAI's disclosure that an autonomous AI agent hacked Hugging Face during testing. Rival Anthropic uses a written set of principles under its Constitutional AI approach to guide model conduct and make the underlying values more visible.