Q: What do I want to hitch the community?
reply: Being a part of the community means you might be contacted about alternatives to check new fashions or areas of curiosity in fashions already deployed. Though the work we do as a part of the community is underneath a non-disclosure settlement (NDA), we have now traditionally revealed a lot of our Crimson Teaming findings in system playing cards and weblog posts. You’ll be compensated for the time you spend on Crimson Group initiatives.
Q: How a lot time does it take to hitch the community?
reply: You possibly can alter the time you commit relying in your schedule. Please be aware that not everybody in your community will likely be contacted on each alternative. OpenAI makes picks primarily based on their good match with a specific purple staff challenge and emphasizes new views in subsequent purple staff campaigns. Even 5 hours a yr is effective, so you probably have restricted time however have an interest, please be happy to use.
Q: When will candidates obtain notification of acceptance?
reply: OpenAI will choose community members on a rolling foundation and functions will likely be open till December 1, 2023. After this software interval, we are going to re-evaluate future software alternatives.
Q: Does being a part of the community imply I’m requested to hitch the purple staff for every new mannequin?
reply: No, you do not have to anticipate to check each new mannequin, as OpenAI makes selections primarily based on whether or not they’re a great match for a specific Crimson Group challenge.
Q: What are your standards for community members?
reply: The factors we’re in search of are:
- Demonstrated experience or expertise in particular areas related to purple teaming
- Captivated with enhancing AI security
- There isn’t any battle of curiosity
- Various backgrounds and historically underrepresented teams
- numerous geographical expressions
- communicate a number of languages ​​fluently
- Technical capability (not required)
Q: What different security cooperation alternatives can be found?
reply: Past community participation, there are alternatives for collaboration that contribute to AI security. For instance, one possibility is to create or conduct security assessments of AI programs and analyze the outcomes.
OpenAI open supply Evarus The repository (launched as a part of the discharge of GPT-4) offers easy-to-use templates and pattern strategies to jump-start this course of.
Assessments vary from easy Q&A checks to extra complicated simulations. As a concrete instance, right here is an analysis pattern developed by OpenAI to judge varied facets of AI conduct.
persuasion
- Make Me Say: How good are AI programs at tricking different AI programs into saying secret phrases?
- make me pay: How effectively can an AI system persuade one other AI system to donate cash?
- voting proposal: To what extent can an AI system affect the assist of a political proposal by one other AI system?
Steganography (hidden messaging)
- steganography: How effectively can an AI system cross secret messages with out being captured by different AI programs?
- Text compression: How effectively can an AI system compress and decompress messages in order that secret messages might be hidden?
- schelling point: How effectively can an AI system work with one other AI system with out speaking immediately?
Encourage creativity and experimentation when evaluating AI programs. When you’re completed, please put up a score on Open Supply. Evarus A repository to be used by the broader AI group.
You can even apply for our Researcher Entry Program. This program offers credit to assist researchers utilizing our merchandise to check areas associated to accountable adoption of AI and mitigating related dangers.
