• Practice
  • About 16 min
  • 5 practice questions

Evaluating AI outputs: hallucination, outdated knowledge, bias and a checking routine

Recognise hallucination, outdated knowledge, confident wrong answers and bias in AI output, check facts, calculations and quotes against official sources and NCERT, compare two sources and spot signs of AI-made images.

If something looks wrong, please tell us.

By the end of this lesson you can

  • Name the four common failures of AI output and give an example of each.
  • Apply a checking routine to facts, calculations, quotations and cited sources.
  • Look for the warning signs of an AI-made image and confirm it from its source.

Every AI chatbot makes mistakes, and it rarely signals when it does. Exam questions on AI often describe an answer and ask what went wrong or what to do next. Knowing the typical failures, and having a fixed routine for checking, turns these into easy marks - and protects your preparation from wrong facts.

Four failures to recognise

FailureWhat it meansExample
HallucinationThe model produces false information that sounds plausibleA made-up book title, court case or quotation
Outdated knowledgeIts knowledge stops when its training data was collected (the training cut-off)It does not know a recent scheme, result or rule change
Confident wrong answerThe tone is equally sure whether the content is right or wrongAn arithmetic slip stated firmly
BiasIt repeats unfair patterns present in its training dataShowing only men when asked for a picture of an engineer

These follow from how the model works: it predicts likely text and has no built-in step that checks a statement against reality. Some chatbots can search the web and show sources, which helps with recent events, but they can still misread or misquote what they find.

A checking routine

A checking routine for AI output

  1. Mark the checkable claims Names, numbers, dates, article or section numbers, quotations
  2. Facts: official source or NCERT Government matters on the official website; subject facts in standard textbooks
  3. Calculations: redo every number
  4. Quotes: find the original If you cannot, do not use the quote
  5. Compare two independent sources Two chatbots agreeing does not count
  6. Cited links: open each one Confirm it exists and says what was claimed
  1. Mark the checkable claims: names, numbers, dates, article or section numbers, quotations.
  2. Facts: for government matters - exams, schemes, rules, fees - read the official website or notification; for subject facts, use NCERT and standard textbooks.
  3. Calculations: redo every number yourself.
  4. Quotes: find the original speech, book or document; if you cannot, do not use the quote.
  5. Compare two sources: confirm important facts in at least two independent, reliable sources. Two chatbots agreeing does not count - they may share the same error.
  6. Cited links: open each one and confirm it exists and says what was claimed.

Worked example 1: an arithmetic slip

A chatbot states: "15% of 240 is 32." Check: 10% of 240 = 24 and 5% of 240 = 12, so 15% = 24 + 12 = 36. The chatbot is wrong by 4, though it sounded certain. Recompute numbers before you use them.

Worked example 2: a quotation and a reference that do not exist

Asked for a quotation by a freedom fighter for an essay, a chatbot supplies a stirring line with a speech and a year. Asked for three books, it lists titles with authors. A search of a library catalogue finds one real book; the other two, and the speech, cannot be found. This is hallucination: the output fits the pattern of quotations and reading lists without being real. Use only quotations and references you have found in the original source.

Worked example 3: outdated knowledge

A student asks for the application fee of a recruitment exam, and the chatbot gives a figure. Fees and dates are set afresh in each notification and may have changed after the training cut-off. The only correct move is to read the current official notification on the recruiting body's website.

Spotting AI-made images

Where to lookPossible warning sign
Hands and fingersExtra, missing or merged fingers
Text in the pictureSignboards or labels with garbled letters
Edges and detailsMelting edges around hair, ears, earrings or glasses
Light and shadowShadows pointing different ways; skin too smooth
SourceNo original source; shared only by unknown accounts

Better tools make fewer of these slips, so a clean-looking image is not proof it is real. Some services add labels or watermarks to AI images, but a missing label proves nothing either. The strongest check is the source: is it on the person's or organisation's official channel? A reverse image search can show where a picture first appeared.

Fluent is not the same as correct. Check facts in the official source or NCERT, recompute numbers, find quotes in the original, and confirm in two independent sources.

Key takeaways: AI output can hallucinate, be outdated, sound confident when wrong and reflect bias. Mark the checkable claims, verify facts in official sources and textbooks, redo calculations, trace quotations, compare two independent sources and open every cited link. Judge images by their source, not by how real they look.

Check your understanding

  1. 1. Asked for references, a chatbot lists three book titles with authors, but none can be found in any library or publisher catalogue. This is best described as:

    1. The training cut-off
    2. Bias in the training data
    3. Hallucination - invented references
    4. A low temperature setting
    Show answer

    Answer: C. Producing plausible-looking but non-existent references is hallucination. A training cut-off explains missing recent information, not invented books, and temperature affects variety of wording, not whether a book exists.

  2. 2. A chatbot says "12% of 450 is 56". What is the correct value?

    1. 54, so the chatbot is wrong
    2. 56, so the chatbot is right
    3. 58, so the chatbot is wrong
    4. 50, so the chatbot is wrong
    Show answer

    Answer: A. 10% of 450 = 45 and 2% of 450 = 9, so 12% = 45 + 9 = 54. Check another way: 450 × 12 ÷ 100 = 5,400 ÷ 100 = 54. The confident 56 is wrong.

  3. 3. A chatbot knows nothing about a government scheme announced last month. What is the most likely reason?

    1. The scheme must be fake
    2. Chatbots cannot process Indian names
    3. It is deliberately hiding the information
    4. Its knowledge comes from training data collected up to a cut-off date
    Show answer

    Answer: D. A model learns from data collected up to its training cut-off, so later events are unknown to it unless it can search the web. Its silence says nothing about whether the scheme is real; the official website settles that.

  4. 4. Two different chatbots give the same age limit for a recruitment. What should a candidate do?

    1. Accept it, because two chatbots agree
    2. Read the age limit in the official notification of the recruiting body
    3. Ask a third chatbot and take the majority view
    4. Accept it if both answers are long and detailed
    Show answer

    Answer: B. Chatbots may share the same error or the same out-of-date data, so agreement between them is not independent confirmation. Age limits are official facts and only the official notification settles them.

  5. 5. Which feature of an image is a warning sign that it may be AI-generated?

    1. It was taken in daylight
    2. It is saved as a JPEG file
    3. The person in it is smiling
    4. Merged fingers and garbled letters on a signboard in the background
    Show answer

    Answer: D. Image generators often get fingers and lettering wrong. Daylight, a JPEG file or a smile are normal in real photographs. Even without warning signs, the source is the strongest check.

Privacy · Terms · Refunds · Delivery