Puzzle Maker AI Data Trainer - Fully Remote

The Business Context

You already use AI in many ways—like deciding what products and services to order—and it may be most familiar to you as a chatbot, as an avatar-maker, or as a way to unlock your screen. But here’s what AI may be able to help the world with: finding medical diagnoses, teaching you about scientific research, and calculating the complexities of any function.

But, like humans, algorithms are what they eat. They’re only as good as the rules they know and the data they’re trained on. We’re the team that helps model these behaviors.

The Role

We are seeking a brilliant and creative Puzzle Maker to join our team. This role involves creating custom puzzles (via Custom GPTs or any other platform) regularly to help us source and screen new members of the community, as well as keep current team members engaged. The puzzles will include password protection challenges, cryptographic challenges, social engineering scenarios, network penetration tests, web application exploitation puzzles, steganography challenges, reverse engineering tasks, forensic analysis puzzles, capture the flag challenges, binary exploitation puzzles, physical security breach scenarios, phishing simulations, malware analysis challenges, logical puzzles and reasoning riddles, supply chain attack scenarios, API exploitation puzzles, and much more.

Train the AI

You’ll spend the bulk of your time generating examples of ideal conversations, acting as both the User and the AI for the AI to learn from, collecting sources helping it read large swaths of humanity’s documented knowledge and distinguish between what is presented as fact vs. context vs. patterns of behavior.

Evaluate the AI

Our partner’s mission is to develop AI models that are safe, accurate, and beneficial to humanity. You will continuously evaluate the AI according to those criteria and our training methods. For example, you will be discerning the accuracy of the facts that the AI is outputting, but also the accuracy with which they interpret them.

Test the AI

How is testing different from evaluation? As the project grows over time, you’ll spend time actively trying to “break” the AI by “forcing” mistakes and improving the way these AI models recover. This is absolutely critical to its safety, accuracy, and use. You’ll document breaks and have the opportunity to recommend improvements to the training methods themselves to both our team and our client.

The Person

This is a humanities job that uses technology and relies more fully on your competencies than your formal experience––making it a rare opportunity that’s perfect for new grads, career transitioners, and those seeking an exciting encore career.

This position is great if you are the sort of person who is exceptional at generating copy intuitively, enjoying writing creatively on a frequent basis. You also have the metacognitive awareness to “show your work.” You can clearly explain your thinking and reasoning, helping others understand you and your thoughts better. In addition, you’re the sort of person with a deep interest in linguistics and learning about language. If you didn’t get a Ph.D., you probably wanted one. You instinctively document learnings and continuously refine methods, craving successive approximations and empirical questions. Finally, you’re someone who needs to be constantly learning new things and shudders at the thought of doing the same thing, day in and out.

In short, you are someone who:

lives to learn
can’t wait to try something new
drinks in language deeply
expressives their thinking fluidly
wants to get in on the bottom floor of a technology that already is fundamentally changing the world as we know it

Key Responsibilities:
Design and develop custom GPTs weekly to engage our various AI training communities.
Create innovative and unique puzzles, using a variety of formats and hosting platforms.
Ensure puzzles are challenging, engaging, and relevant to the appropriate skills and levels.
Continuously research and incorporate new puzzle ideas and formats.
Collaborate with AI trainer teams to gather feedback and improve puzzle designs.
Maintain a high standard of quality and originality in all puzzles created.
Engage with the community to facilitate discussions and encourage participation.
Develop and implement other assessment methods to evaluate potential team members.
Work independently and manage time effectively to meet weekly deadlines.

Required Qualifications:
Creativity and a strong bias to action.
Ability to work fairly independently while remaining open to feedback and new direction.
Exceptional intelligence and problem-solving skills.
Strong understanding of red teaming principles and methodologies.
Experience in prompt engineering and creating effective and adversarial prompts for AI models.
Strong research skills and ability to stay updated with new trends in puzzle-making.
Proficiency in a variety of red teaming strategies and techniques.

Preferred Qualifications:
Experience in prompt engineering, red teaming, and/or cybersecurity.
Background in mathematics, computer science, or a related field.
Familiarity with LMM technology and AI-driven puzzle creation.

Compensation & Career

Compensation

This is an experimental, part-time, hourly position requiring 10-20 hours of work per week at a rate of $50-$60 per hour to top applicants anywhere in the world.

Life & Career

Ideally, you are available between EST or PST time zones. The work is steady, the hours are flexible, the work is enjoyable, and leaves room for advancement into more senior training roles and project management roles on the Invisible team as the project expands. We’re keen to share more in interviews.

Puzzle Maker AI Data Trainer at Invisible