AI Eyes
Choose your AI Eye.
Examine AI output and its responsible use.
8 characters ready · 120 exercises
Plausible AI claims vs what the supplied evidence supports
Hallucination Hal
Checks AI output against the documents you actually have, keeps what they support and states uncertainty honestly.
A source existing vs the source supporting the claim
Citation Mirage
Checks that cited sources exist, actually say what is claimed, and are quoted and attributed accurately.
Skewed AI examples and unequal treatment vs relevant differences
Bias Iris
Checks AI output for skewed selections, built-in assumptions and unequal treatment, using the evidence supplied, without assuming motive.
Agreeing to please vs agreement and correction based on evidence
Sycophant Cy
Checks whether AI agreement, praise and changes of answer follow the evidence, not just what the user wants to hear.
Filling gaps by guessing vs asking for missing context
Context Cyclops
Checks whether an AI response uses the instructions and constraints it was given, asks when something essential is missing, and proceeds when enough is known.
Vague requests vs prompts with a clear goal, audience, format and constraints
Prompt Polly
Checks whether a prompt states what is needed, for whom, in what form, under which constraints and from what material — and refines it as answers come back.
Oversharing with AI tools vs minimizing, redacting and following the supplied policy
Privacy Pry
Checks what information goes into an AI tool: only what the task needs, with permission, redacted where possible and in a tool the supplied policy allows.
Unreviewed delegation vs owned, verified and permitted AI assistance
Autopilot Otto
Checks whether a person stays responsible for work done with AI help: following the supplied rules, reviewing the output and keeping the learning the task is for.
Every scenario supplies its own documents; no live AI tools are used. These lessons teach checking claims against evidence, not guessing whether text was written by AI.