Chain-of-Thought Prompting: Ask for Checkable Work

What does chain-of-thought prompting mean?

The research term describes prompting a language model to produce intermediate reasoning steps before a final answer, often through worked demonstrations. Related zero-shot research tested step-by-step phrasing without demonstrations. Results depend on the model, task and setup; a phrase does not guarantee correctness.

• Wei and colleagues: worked-demonstration studyLearn more
• Kojima and colleagues: zero-shot studyLearn more

Ask for work you can verify

Request a concise explanation, stated assumptions, relevant calculation or test and a final answer. This gives you material to check. It does not reveal or certify a model’s private internal reasoning. For a simple formatting task, extra explanatory text may make the result less useful.

A budget you can calculate yourself

Use this practice request, then compare its answer with the manually checked values below. Tax and delivery are unknown and stay separate.

1Using only this supply plan, return a compact calculation table, total and remaining budget: 2 notebooks at USD 15 each, 3 marker packs at USD 8 each, budget USD 100. Show the arithmetic and units briefly. Tax and delivery are unspecified; keep them separate. I will check each multiplication and subtraction independently.

Verify calculations and claims independently

Recompute the arithmetic, check units and source inputs, and inspect whether an assumption became a fact. For a code result, run relevant tests; for research, read actual sources. More words or a plausible explanation do not replace these checks.

Explanations can sound convincing and still mislead

A study found that tested models could rationalize answers influenced by biased prompts without acknowledging that influence. A visible explanation is useful material to evaluate, not a guaranteed faithful account of the computation that produced the answer.

• Turpin and colleagues: explanation faithfulness studyLearn more

See formulas in an actual spreadsheet

For a larger example, inspect the existing Work budget and printing revision. Check its source inputs, native formulas and revised totals in the spreadsheet lesson.

• Inspect the actual workbook and revisionLearn more

When a step is wrong, target the error

Name the wrong input, operation or unsupported assumption and supply the correct source. Ask for a corrected result and concise check, then recompute it. Avoid accepting a different final number merely because a second explanation sounds better.

1The marker-pack quantity is 3 and unit cost is USD 8. Recompute the known subtotal and remaining USD 100 budget from those supplied inputs. Preserve the notebook figures. Keep tax/delivery unspecified and show the two calculations briefly.

Trying these tasks in Gab AI

Open Chat with a model available to your account. Models and Credits explain request access and usage.

• Models and request accessLearn more

Reasoning-prompt questions

Ask for inspectable work and check it.

Try the prompt in Chat

Copy the prompt, open Chat and paste it into the composer. Read the answer beside the supplied facts before requesting the revision below.

These buttons open the product and copy your draft. You choose the mode and send the request yourself.

Make one change and check the whole result

Review the first answer, then send this follow-up in the same conversation.

1Change marker packs from 3 to 4 at USD 8 each. Preserve 2 notebooks at USD 15 each and the USD 100 budget. Show the corrected subtotal and remaining budget with units. Keep tax and delivery unspecified. Return a concise calculation table and final answer.

Check that the requested change happened and the correct facts, format and missing details survived. For a file, open the revised artifact and inspect nearby paragraphs or formulas too.