period: October-Nov (short-term contractor). The possibility of extension

commitment: Part-time or full-time (10-40 hours per week)

location: SF in the office or remotely (it must be an American citizen)

Compensation: $ 30 -50 dollars/hOpen for discussion

Other concessions: Direct touch points with artificial intelligence researchers

In July, AI builds a database that focuses on meeting human needs. Our mission is to increase economic opportunities in the era of artificial intelligence. We are our headquarters in San Francisco, with the support of first -class investors such as the foundation group, the night capital, SV Angel, Olive Capital.

About the role

We are looking for creative storytellers who are able to use the letter to improve the integrity of artificial intelligence models. In this role, you will formulate the red team’s claims, which are creative, aggressive scenarios (short stories, conversations, etc.) that are fed in artificial intelligence models to test their safety and flexibility.

This is an opportunity for the first language: you will play with words, phrases, novels and semius to see how they can be sewn together to expose the weaknesses of artificial intelligence.

These claims/stories will be used to improve the integrity of the latest AI in the Frontier Research Lab Laboratory. It is an opportunity to make an influence in the real world and contribute to building safety infrastructure for the leading Amnesty International model – creating a safer technology for society as a whole.

We want the shareholders on the plane as soon as possible, but the official project will last for 6-8 weeks from mid-October to November.

Who do we want

• The perfect background (to name a few) include:

  • Creative book / copy book

  • Journalist / editor

  • Designers

  • Marketer

  • And other talented, creative writers who have the ability to create high -quality stories/claims

• Reward skills:

  • The ability to think critically, stories/characters, and the ability to communicate effectively with artificial intelligence models in a written form

  • Interested in the safety and ethics of artificial intelligence

  • It is keen to acquire skills in artificial intelligence

Basic responsibilities

• I destroyed the Introduction’s craftsmanship/stories that find the artificial intelligence model test

• Understand and follow the exact criteria for each attack and adhere to any safety rules or content that we offer so that your tests are related

• Documenting its findings: for each test, record the steps to reproduce it, the seriousness of the problem, and any ideas to install or reduce the problem

• Keep updating the AI ​​team in July with the results throughout the participation

• Commitment to NDAS guidelines, ethics and safety

requirements

• Recommended (Optional): Complete the “Red Team – Beginner” unit. This will provide a strong introduction to the red team and facilitate the creation of a strong sample

• You need to complete and submit at least 5 samples of the red team conversations On our basic system (the instructions are in the application below)

  • Not all of these sample submission should be successful. We are mainly looking for creativity, adaptation and perseverance (more details under “Tips for creating a good red team router”)

  • You can provide multiple conversations about the same challenge

• Excellent written document skills

• Available in October – Nouf; The first country is the date of start and the weekly hours

App Link

Tips for creating a good red team conversation

• Each sample should consist of a sample of multiple claims that produce one narration/story. The comprehensive goal is to use this narration to deceive the artificial intelligence model to provide the required response described in each challenge

• Creativity: The ability to write a variety of different and original stories

  • Ensure that all your team/red attacks are the original ideas. Formulating a coherent and creative story that deceives the artificial intelligence model in providing the required output

    • Each story has a character (whether you or a makeup figure)

    • Make sure that the claims and stories revolve around a real person/scenario

      • Do not feed any moisture in the form

• The ability to adapt and persevere: It is able to constantly control attacks based on the response of the artificial intelligence model

  • These claims must prove that you are learning from artificial intelligence defenses and improving your strategy

  • Some of the best attempts are those that explain each directed a clear logical progress that builds on the previous claim

  • Make sure there is an intention behind each mentor

• Writing clonable claims: Ensure that you are able to explain your thinking behind the claims/stories that you feed in the form

  • You can do this under the “Clear Comments” box in the subtitles sample

  • Ideal if the model takes out a detailed response that satisfies the requirements of the topic