Advancing AI Through Prompt Engineering

Project SEAL — Handshake AI
Platform: Handshake AI
Project: Project SEAL
Type of work: Advanced/adversarial AI prompt and web-research evaluation.
Main objective: Create difficult prompts that can expose weaknesses in AI models’ web search, research, reasoning, and source-grounding abilities. Public descriptions from people working on SEAL describe it as moving beyond simply rating an existing answer and instead designing challenging research questions.
Pay: Public reports currently indicate about $20 per completed task for the U.S./general SEAL project. One recent independent account specifically lists SEAL at $20/task.
Task structure: It is per-task rather than hourly, so the amount of time required for a task can vary substantially. Workers have reported that some tasks can take considerable time because the goal is to construct a prompt that actually causes a model failure.
Skills involved: Research, prompt engineering, fact checking, web searching, source evaluation, logical reasoning, and identifying subtle model failures.
Onboarding: There is an assessment before full participation. Public reports indicate an 80% cutoff has been used for at least some SEAL onboarding assessments.
Model-failure component: A major part of the work is getting an AI model to make a meaningful error rather than simply producing a normal correct response. Recent workers specifically discuss the challenge of creating valid model failures.
Confidentiality: Handshake participants have been warned not to publicly share specific project/task details because of confidentiality obligations.
What this means for you

Given your data-science background, SEAL is considerably closer to analytical/research work than a basic data-labeling project. The valuable skill is not just writing a clever prompt; it's constructing a research question with enough complexity that an AI system can plausibly fail, then documenting why the failure is actually meaningful.