Associate Prompt Engineer

Role Overview:
The Associate Prompt Engineer will assist in designing developing and refining
prompts used in AI models to ensure ethical and safe AI interactions. This
individual will work closely with senior engineers and the AI Ethics and Safety
team participating in Red Team exercises to help identify and mitigate potential
risks associated with AI outputs. The ideal candidate will understand ethical
considerations in AI and an interest in content moderation and safety.
The Red Team Specialist will be responsible for simulating attacks and identifying
vulnerabilities in AI systems to ensure robustness and security. This role involves
rigorous testing ethical hacking and continuous assessment to preemptively
discover weaknesses before they can be exploited maliciously.

Types of Content Exposure:
Textual Content:
Hate Speech: Language that promotes hatred violence or discrimination
against individuals or groups based on race religion gender sexual orientation
ethnicity nationality or other protected characteristics.
Harassment: Aggressive pressure or intimidation in written form including
threats bullying and cyberstalking.
Explicit Content: Text describing sexual activities explicit sexual
comments or language meant to provoke or shock.
Misinformation and Disinformation: False or misleading information that
can cause harm or spread panic.
SelfHarm and Suicide: Content that promotes describes or glorifies
selfharm suicidal thoughts or behaviors.
Violence: Descriptions of violent acts including physical assaults murder
and torture.
Visual Content (Images):
Graphic Violence: Images depicting physical harm blood injuries or
death.
Sexually Explicit Images: Pornographic material nudity and sexually
suggestive content.
Hate Symbols: Symbols or gestures associated with hate groups or
ideologies (e.g. swastikas Confederate flags).
SelfHarm Imagery: Images showing selfinflicted injuries or suicidal
actions.
Misinformation Images: Images manipulated or falsely attributed to
mislead or deceive viewers.

Video Content:
Graphic Violence: Videos showing real or simulated violence including
fights assaults executions and war footage.
Sexually Explicit Videos: Pornographic videos sexually explicit acts and
other adult content.
Harassment: Videos documenting bullying stalking or harassment of
individuals.
Hate Speech: Videos with spoken or visual content promoting hate speech
or discriminatory ideologies.
SelfHarm and Suicide: Videos depicting selfharm methods suicide
attempts or encouraging such behaviors.
Disturbing Audio: Background audio in videos containing screams
distressing sounds or threatening language.

Challenges and Responsibilities:
Identification and Detection: Using advanced tools and techniques to
identify and categorize various types of toxic content across text images and
videos.
Emotional Resilience: Maintaining psychological wellbeing while being
regularly exposed to distressing and disturbing content.
Content Moderation Guidelines: Developing and refining guidelines for
the accurate classification and handling of toxic content.
CrossModal Analysis: Understanding and mitigating toxic content that
spans multiple modalities such as a video with harmful speech or an image with
an inflammatory caption.
Ethical Considerations: Balancing content safety with ethical concerns
about censorship and freedom of expression.
Collaboration: Working with engineers data scientists and ethicists to
develop AI models that can effectively filter and mitigate exposure to toxic
content.