Refining and Regenerating Prompts – a ChatGPT tip from Ray Poynter

Refine or Regnerate - TipPublished by Ray Poynter

In this post, I will highlight two approaches to getting to a better answer from ChatGPT.

Refining
Most times, when I am trying to get a result or answer from ChatGPT, I will use some level of refining (or iterating) the question. I ask a question, then I ask a follow-up question that builds on the previous one.

For example, I might ask ChatGPT “If I have a sample of 1000 people and I find that 50% of people like the test item, what is the confidence interval at the 95% level of confidence?

ChatGPT replied by telling me all about the formula for this test and produced an answer of CI=0.5±0.031.

The first refinement might be a simple tidying up of the answer.

I could ask “Write the CI = line using percentages instead of decimals.”

ChatGPT replied with CI=50%±3.1%

But I might have a more refined question that I need to ask.

For example, I might ask ChatGPT “The population size for this study was 10,000, does that change the CI?”

ChatGPT tells me that the finite population correction factor needs to be applied and that the answer is now CI=50%±2.94%.

This refinement process should be your norm. Ask a first question, then keep refining the question for get to the question you ought to ask.

Regeneration
Regeneration is Different from Refinement. Regeneration is a request to repeat the last question to see if an alternative answer is better.

For example, I might ask ChatGPT, “I have developed a new platform that conducts driver analyses for clients, such as hotels, restaurants, car hire, and travel. Suggest five possible names for use in Western Europe.”

ChatGPT generated five names. If I was not very impressed, I could Refine my question. Or, I could click on the symbol highlighted in the image below (the curved arrows facing the opposite way) and it will generate five new names.

The Regenerate symbol

When I am asking ChatGPT to generate names, to cluster words into themes, or to generate synthetic responses, I find the Regenerate option can be a useful and quick alternative to the Refinement route.

Curious about the names?

The five names it gave me the first time I asked were:

  1. InsightDrive
  2. DriverMetrics
  3. SatisFactor
  4. RevelaMetrics
  5. PerformaInsight

When I pressed the Regenerate symbol I got:

  1. DriverKey
  2. InsightPulse
  3. FactorFocus
  4. TrendDriver
  5. PeakInsights