Overview of the Anthropic and Accenture Partnership
Leading Model Evaluations
- Facilitated by Faculty: Accenture’s specialist AI business, Faculty, will be responsible for leading the evaluations of AI models.
- Funding and Collaboration: Anthropic will fund Accenture’s evaluations while exploring collaborative pilots with groups like the nonprofit evaluator METR.
Non-Exclusive Partnership
- Broad Collaboration: The partnership is non-exclusive; Anthropic intends to include more independent evaluators as it continues to engage with other AI developers.
Investment Plans
- Substantial Investments: Both Anthropic and Accenture plan to invest at least $1 billion each over the next five years into building evaluation capabilities for frontier AI models, aiming to improve model safety and alignment.
Embedded Evaluation Strategy
- Independent Evaluators Inside Companies: The partnership promotes the integration of independent evaluators within AI companies, allowing for more comprehensive oversight.
- Direct Engagement: Embedded evaluators will have access comparable to employees, enabling them to observe model training, decision-making processes, and communicate directly with staff.
Funding Framework
- Long-term Funding Goals: Anthropic advocates for future funding to come from pooled or government sources, as delineated in its Advanced AI Framework. Current efforts involve different evaluators funded through various arrangements.
Future Directions and Evolutions
- Ongoing Innovation: Anthropic is committed to continually training and releasing frontier models, iterating on its evaluative strategies as the field evolves. Further adjustments and partnerships with additional evaluators are anticipated as the initiative develops.
