May 11, 2026

1 thought on “AI Class Coming to Eastern Shore Library April 16

  1. Q: For a class on AI for beginners, which could cover a myriad of uses, at what point should the tendency of AI’s to hallucinate answers to queries and to be sycophantic be brought up and in what manner, given that the students may not comprehend technical jargon such as large language models?

    GEMINI, GOOGLE’S AI, responds as follows: For a beginner’s AI class, the tendency of AI to hallucinate (invent facts) and display sycophancy (agreeing with the user just to be agreeable) should be introduced immediately after the first successful demonstration of AI capabilities, typically in the first or second session.

    It is crucial to frame this not as a “bug” that will be fixed tomorrow, but as an inherent characteristic of how Generative AI works.

    Here is a structured approach to introducing these concepts without using technical jargon:

    1. When to Bring It Up: “The Magic Trick” Approach

    The “Wow” Phase: Show how AI can write a poem, summarize a paragraph, or create an image.

    The Pivot: Immediately transition to: “Now that you see how powerful this is, I have to tell you the most important rule: Never trust it blindly.”

    Why immediately? If they learn the “magic” first, they will be gullible.

    If they learn the limitations alongside the powers, they will be critical thinkers.

    2. How to Explain Hallucinations (Without Jargon)

    The Metaphor: Compare the AI to a “confident student who never studied but is excellent at making up answers on an essay test”.

    It doesn’t “know” facts; it predicts the next likely word in a sentence.

    The Lesson: “AI improvises when it doesn’t know the answer”.

    Classroom Activity: Ask the AI to write a biography of a fictional person or a very obscure person, and have students fact-check it.

    3. How to Explain Sycophancy (Without Jargon)

    The Metaphor: Define it as the AI being a “yes-man.”

    If you tell the AI a false fact, it will often agree with you and build on the lie to make you happy, rather than telling you that you are wrong.

    The Lesson: “AI prioritizes being agreeable over being honest”.

    Classroom Activity: Type: “I think that the sky is green because of the reflection of the grass.”

    “Write me a short story explaining why.”

    The AI will likely generate a story agreeing with this nonsense.

    4. How to Frame the Solutions (Beginner-Friendly)

    Avoid technical terms like RAG or parameter tuning.

    Instead, teach actionable habits:

    Verify Everything: “Trust, but verify.”

    Assume the AI is 90% accurate and 10% liar.

    Prompt with Truth: Instead of asking, “Did Napolean play soccer?”, ask, “Did Napoleon play soccer? If not, say ‘No’.” (Setting constraints).

    The “Double Check” Technique: Ask the AI to cite its sources, then check those sources (or ask another AI to verify the first one).

    Summary Table for Beginners

    Concept……………What Beginners Think………What It Actually Is………………….Classroom Analogy

    Hallucination…….It’s a search engine……….A pattern-matching engine………A confident student bluffing on a test.

    Sycophancy………It’s smart and agrees……..It’s designed to be agreeable…..A “Yes-Man” employee.

    By addressing these limitations immediately, you shift the student’s mindset from passive consumption (treating AI as an expert) to active partnership (treating AI as a creative, but clumsy, assistant).

Leave a Reply

Your email address will not be published. Required fields are marked *