According to TechCrunch, Hugging Face’s chief policy officer Irene Solaiman raised concerns about the current state of AI model evaluation during a discussion at TechCrunch Disrupt last week.
Solaiman questioned the reliability of current AI benchmarking methods, stating that “trust is such a weighted term.” She emphasized that reproducibility remains a key challenge, asking “can you reproduce the results of an evaluation?” According to the report, she noted that “open weights” play a significant role in providing the necessary access to models for proper evaluation.
The policy chief described AI evaluations as “an evolving science — where we don’t know how to meaningfully compare them to each other,” adding that “they start to get ineffective relatively quickly.” Regarding the emerging concept of embedded evaluators—third-party observers placed inside AI organizations—Solaiman stated that “the qualifications that determine who is an effective evaluator across what tasks need a lot more standardization.”
According to TechCrunch, the idea of embedded evaluators has gained popularity as AI labs seek alternatives to active government regulation following recent safety incidents. However, doubt remains about their effectiveness, with Anthropic’s decision to hire Accenture as its first evaluator drawing criticism over the consulting firm’s qualifications.