Joining OpenAI’s Network
Are you interested in testing new AI models or exploring specific areas of interest? Joining OpenAI’s network gives you the opportunity to do just that. As a network member, you may be contacted to participate in red teaming projects. These projects are conducted under a non-disclosure agreement (NDA), but some findings may be published in System Cards and blog posts. Plus, you’ll be compensated for your time.
What Will Joining the Network Entail?
If you join OpenAI’s network, you may be selected to test a new model or explore a specific area of interest on an already deployed model. It’s a great chance to contribute your expertise and passion for AI safety. The time commitment is flexible, so it can easily fit into your schedule. Even just 5 hours in a year would be valuable. The selection process is based on the right fit for each project, and new perspectives are highly valued.
When Will You Be Notified of Acceptance?
The selection process for network members is ongoing, and applications are accepted until December 1, 2023. If you’re interested, don’t hesitate to apply. Even if you’re not chosen immediately, future opportunities may arise.
Will You Test Every New Model?
No, being part of the network doesn’t mean you’ll be asked to test every new model. OpenAI carefully selects members based on their fit for specific red teaming projects. So, you should not expect to test every new model.
Criteria for Network Members
OpenAI is looking for network members who have demonstrated expertise or experience in a relevant domain for red teaming. They value individuals who are passionate about improving AI safety, have no conflicts of interest, come from diverse backgrounds and underrepresented groups, have diverse geographic representation, and are fluent in multiple languages. Technical ability is a bonus but not required.
Other Collaborative Safety Opportunities
Joining the network is not the only way to contribute to AI safety. OpenAI offers other collaborative opportunities, such as creating and conducting safety evaluations on AI systems. This involves analyzing the results to ensure AI behaves responsibly.
OpenAI’s open-source Evals repository, released during the GPT-4 launch, provides user-friendly templates and sample methods to kickstart the evaluation process. Evaluations range from simple Q&A tests to more complex simulations. OpenAI has developed sample evaluations that assess AI behaviors from different angles, like persuasion and steganography.
OpenAI encourages creativity in evaluating AI systems and values contributions from the broader AI community. Once you complete your evaluation, you can contribute it to the open-source Evals repository.
Additionally, you can apply to OpenAI’s Researcher Access Program, which offers credits to support researchers studying the responsible deployment and risk mitigation of AI.