This job post has expired on September 10, 2024. It is likely that the position has already been filled.
We're at the forefront of AI development, focusing on creating safe, accurate, and beneficial AI models. Our work involves training AI systems to handle complex tasks across various fields, from medical diagnostics to scientific research.
We're seeking a creative Puzzle Maker to join our AI training team. This part-time, remote position involves:
- Creating diverse, challenging puzzles for AI training and team engagement
- Generating ideal AI-human conversations for training purposes
- Evaluating AI performance for safety, accuracy, and beneficial outcomes
- Testing AI models to identify and address weaknesses
- Design weekly custom GPTs and puzzles (e.g., cryptographic challenges, network penetration tests, capture the flag challenges)
- Train AI models using conversation examples and knowledge sources
- Evaluate AI outputs for factual accuracy and interpretation
- Conduct "break testing" to improve AI resilience and safety
- Document findings and suggest improvements to training methods
We're looking for someone who:
- Is passionate about learning and innovation
- Has strong linguistic skills and creative writing abilities
- Can clearly articulate thought processes
- Adapts quickly to new concepts and technologies
- Has a background or strong interest in humanities, combined with technological aptitude
- Exceptional creativity and problem-solving skills
- Strong understanding of red teaming principles
- Experience in prompt engineering
- Ability to work independently while being receptive to feedback
- Excellent research and analytical skills
- Background in cybersecurity, mathematics, or computer science
- Familiarity with LMM technology and AI-driven puzzle creation
- $50-$60 per hour for top applicants
- 10-20 hours per week
- Flexible schedule, preferably available during EST or PST business hours
Opportunities for advancement into senior training roles and project management positions as the project expands.