Understand why AI can sound right and be wrong
https://patrickz.ai/learn/understand-how-ai-answers/

OUTCOME
See how context changes a generated answer and practice checking one claim.

YOU NEED
Any chat assistant and a short fictional note. No coding required.

1. Observe a continuation
Ask the assistant to continue “The bank was…” in a river scene and then in a finance scene. Compare the words that become plausible when context changes.

2. Distinguish generation from verification
Give a fictional note containing a person, task and deadline. Ask a question the note cannot answer. Notice whether the assistant admits the gap or fills it.

3. Add a grounding rule
Repeat with an instruction to use only the note, quote support and say unknown when information is missing. This helps make errors visible but does not guarantee correctness.

4. Check the output yourself
Trace one answer to the source sentence. For real-world claims, open a trustworthy source and compare. Keep uncertainty when the evidence is absent.

PROMPT
Answer using only this note: [note]. For each answer quote the supporting phrase. If the note does not contain the answer, say “unknown from this note.” Questions: [questions]. Do not turn a plausible guess into a fact.

EXAMPLE / EXPERIMENT
Note: “Maya will draft the outline.” Question: “When is it due?” The supported answer is unknown. Friday may sound sensible, but the note never said it.

CHECK YOUR RESULT
[ ] You can explain how context changed the continuation.
[ ] A missing fact remains unknown.
[ ] You independently checked one supported answer.

IF IT FAILS
If the assistant invents support, compare the quote character by character with the input. Tool use or search can supply evidence, but you still need to inspect the source.

MY RESULT / NEXT CHANGE