According to OpenAI, the board of directors has the final say on the safety of new AI models as it aims to mitigate "catastrophic risks."

OpenAI, the company behind viral chatbot ChatGPT, has published a 27-page document detailing their approach to mitigating "catastrophic risks" through modern AI models. This "Preparedness Framework" outlines how the company will monitor, evaluate, and safeguard against potential threats, with the board of directors holding the final say on the release of new AI models.

The board of directors, including three affluent white men, will ensure that the high-tech AI tools developed by OpenAI are utilized for the betterment of humanity. Critics have raised concerns about the lack of diversity and the reliance on self-regulation, prompting calls for greater legislative involvement in the safe development and application of AI technology.

OpenAI's strategy involves both robust safety measures and strategic decision-making. This approach is known as "deliberative alignment," which goes beyond standard techniques like RLHF and RLAIF by incorporating a more holistic process to align AI models with desired outcomes.

The company employs various safety strategies, such as moderation models, blocked websites, and real-time moderation systems, to prevent misuse and ensure ethical standards. By involving the board of directors and adhering to global AI policy frameworks, OpenAI strives to develop and deploy responsible AI models, minimizing the risk of catastrophic events.

Source:

Enrichment Data:

1. Deliberative Alignment: OpenAI employs a comprehensive approach to AI safety, known as "deliberative alignment." This method involves aligning AI models with desired outcomes beyond standard approaches like RLHF and RLAIF through a holistic process.
2. Safety Strategy: OpenAI's safety strategy for their o3 model focuses on responsible deployment through rigorous testing and validation, including measures to prevent misuse and adherence to ethical standards.
3. Board Involvement: The board of directors plays a significant role in overseeing the development and deployment of AI models, with co-founder Ilya Sutskever actively participating in discussions about safety and capabilities.
4. Safety Measures: OpenAI implements safety measures such as moderation models, blocked websites, and real-time moderation systems to mitigate risks associated with AI agents like Operator. These measures include user confirmation, website limits, and monitoring systems to prevent unintended actions.
5. Governance Initiatives: OpenAI engages in governance initiatives that align with global AI policy frameworks, including participation in AI safety summits and collaborations with other organizations to establish safety standards.
6. Transparency and Accountability: Despite concerns about transparency and accountability in AI development, OpenAI has taken steps to address these issues, with internal conflicts and external scrutiny leading to discussions about the need for stronger safety measures.