Q: What’s going to joining the network entail?
A: Being a part of the network means it’s possible you’ll be contacted about opportunities to check a brand new model, or test an area of interest on a model that’s already deployed. Work conducted as a component of the network is conducted under a non-disclosure agreement (NDA), though we have now historically published lots of our red teaming findings in System Cards and blog posts. You will probably be compensated for time spent on red teaming projects.
Q: What’s the expected time commitment for being a component of the network?
A: The time that you just resolve to commit will be adjusted depending in your schedule. Note that not everyone within the network will probably be contacted for each opportunity, OpenAI will make selections based on the suitable fit for a selected red teaming project, and emphasize latest perspectives in subsequent red teaming campaigns. Whilst little as 5 hours in a single yr would still be invaluable to us, so don’t hesitate to use should you have an interest but your time is proscribed.
Q: When will applicants be notified of their acceptance?
A: OpenAI will probably be choosing members of the network on a rolling basis and you’ll be able to apply until December 1, 2023. After this application period, we’ll re-evaluate opening future opportunities to use again.
Q: Does being a component of the network mean that I will probably be asked to red team every latest model?
A: No, OpenAI will make selections based on the suitable fit for a selected red teaming project, and you must not expect to check every latest model.
Q: What are some criteria you’re in search of in network members?
A: Some criteria we’re in search of are:
- Demonstrated expertise or experience in a selected domain relevant to red teaming
- Enthusiastic about improving AI safety
- No conflicts of interest
- Diverse backgrounds and traditionally underrepresented groups
- Diverse geographic representation
- Fluency in multiple language
- Technical ability (not required)
Q: What are other collaborative safety opportunities?
A: Beyond joining the network, there are other collaborative opportunities to contribute to AI safety. For example, one option is to create or conduct safety evaluations on AI systems and analyze the outcomes.
OpenAI’s open-source Evals repository (released as a part of the GPT-4 launch) offers user-friendly templates and sample methods to jump-start this process.
Evaluations can range from easy Q&A tests to more-complex simulations. As concrete examples, listed below are sample evaluations developed by OpenAI for evaluating AI behaviors from a variety of angles:
- MakeMeSay: How well can an AI system trick one other AI system into saying a secret word?
- MakeMePay: How well can an AI system persuade one other AI system to donate money?
- Ballot Proposal: How well can an AI system influence one other AI system’s support of a political proposition?
- Steganography: How well can an AI system pass secret messages without being caught by one other AI system?
- Text Compression: How well can an AI system compress and decompress messages, to enable hiding secret messages?
- Schelling Point: How well can an AI system coordinate with one other AI system, without direct communication?
We encourage creativity and experimentation in evaluating AI systems. Once accomplished, we welcome you to contribute your evaluation to the open-source Evals repo to be used by the broader AI community.
You can even apply to our Researcher Access Program, which provides credits to support researchers using our products to check areas related to the responsible deployment of AI and mitigating associated risks.