Signing in

You will be sent to MillerKnoll sign-in.

AI

How AI Gets Things Wrong

AI outputs are wrong more often than most users realize, reliably, in specific ways. Understanding those patterns is the foundation of judgment.

Lesson 1

Predictable patterns, not random errors.

Hallucination: factually incorrect content stated with complete confidence, wrong stats, fake citations, incorrect titles. The model generates plausible continuations; it does not retrieve verified facts.

How AI gets things wrong

HallucinationPlausible fictionConfident toneSounds certainStale dataOld training cut
Hallucination, confident tone, stale data, predictable patterns, not random luck.

Core principles

  1. Most likely wrong on: specific numbers, dates, percentages, proper nouns, recent events, niche topics, precise citations.
  2. Confident wrongness: hesitation and "I think" do not apply, verify consequential claims regardless of tone.
  3. Stale information: training cutoffs and outdated org context, ask if the answer must be current now.
  4. Bias: training data reflects human gaps, notice narrow perspectives in broad questions.

Related, Getting Started: judgment

Check yourself

What does "hallucination" mean in the context of AI outputs?

Do this in Copilot

Pick one claim from your next Copilot output (number, name, date) and verify it independently.

Paste this into Copilot Chat and work through it before moving on.

Surface uncertainty

Summarize this content. For any claim where you are not fully grounded in the provided material, say so explicitly instead of guessing.
Open Copilot →
  • Uncertainty flagging

Did you run this in Copilot? Mark complete when you have tried it.

Next lesson: A Practical Verification Framework →