An update on independent evaluators
Anthropic announced an update regarding how outside groups assess its artificial intelligence systems. The company is granting "employee-like access" to METR, a non-profit entity focused on reviewing frontier models, to assist with safety evaluations. Additionally, Anthropic is engaging in preliminary talks with other entities, including Hugging Face.
Key Takeaways
- Anthropic has announced an initiative to enhance system safety assessments by granting internal-style access to outside experts.
Specifically, a team from METR, a non-profit organization that specializes in examining frontier artificial intelligence models, is receiving "employee-like access" to conduct evaluations.
- Beyond its work with METR, Anthropic revealed that it is holding early discussions with several other entities, including Hugging Face.
Partnering with independent evaluators helps developers gain third-party insights into model risks, offering an example of how safety procedures are structured in industry settings.
- Anthropic is providing "employee-like access" to a team from METR to conduct safety evaluations on its systems.
METR operates as a non-profit organization dedicated to evaluating frontier artificial intelligence models.
- Anthropic is in early discussions with additional organizations, including Hugging Face, for safety evaluation partnerships.
- This arrangement allows external reviewers to scrutinize models with permissions similar to internal staff.
Anthropic has announced an initiative to enhance system safety assessments by granting internal-style access to outside experts. Specifically, a team from METR, a non-profit organization that specializes in examining frontier artificial intelligence models, is receiving "employee-like access" to conduct evaluations. This arrangement allows external reviewers to scrutinize models with permissions similar to internal staff.
Beyond its work with METR, Anthropic revealed that it is holding early discussions with several other entities, including Hugging Face. Partnering with independent evaluators helps developers gain third-party insights into model risks, offering an example of how safety procedures are structured in industry settings. Anthropic is providing "employee-like access" to a team from METR to conduct safety evaluations on its systems.
For more details please read the original article at Anthropic.
Continue Learning
Comments
Comments appear only after moderation. Your email identifies your submission to the moderator and is never displayed here.
No approved comments yet.