Subject: Feature Suggestion: Peer Review Mode
I have a feature suggestion based on how I use ChatGPT professionally.
One of the biggest challenges with AI is not whether it can produce good answers. It is knowing when to trust those answers. While reasoning models continue to improve, what I often need is not simply a recommendation, but a way to understand how much confidence I should place in it and what assumptions are driving the conclusion.
For many professional users, the cost of confidently accepting an incorrect recommendation is much higher than the cost of taking a little longer to arrive at a well-supported one.
I would like to see an optional interaction mode called Peer Review Mode.
The goal would not be to make the model argumentative or automatically disagree with the user. Instead, it would behave like an experienced engineering or technical peer reviewing a design, proposal, or decision before it moves forward.
A good peer reviewer doesn’t try to prove someone wrong. They try to determine whether the conclusion is justified by the available evidence and whether important risks or assumptions have been overlooked.
Some behaviors I would expect include:
- Clearly separating established facts from inferences, assumptions, and opinions.
- Identifying the key assumptions behind a recommendation.
- Presenting the strongest reasonable counterargument or alternative explanation.
- Highlighting missing information that could materially change the conclusion.
- Explaining what evidence would increase or decrease confidence in the recommendation.
- Pointing out potential cognitive biases, such as confirmation bias, anchoring, or overconfidence, when they appear to influence the discussion.
- Assigning an appropriate confidence level to major conclusions instead of presenting everything with equal certainty.
- Explicitly stating when multiple conclusions are equally reasonable based on the available evidence.
- Recommending additional information or analysis before making a decision when appropriate.
I would also find value in an optional Decision Summary that explains the basis for the recommendation without exposing the model’s internal reasoning process. For example, it could summarize:
- The primary alternatives that were evaluated.
- The factors that most influenced the recommendation.
- The assumptions that introduce the greatest uncertainty.
- The information that would most improve confidence in the recommendation.
As a project manager, I don’t use ChatGPT primarily to validate decisions I’ve already made. I use it as a technical reviewer, consultant, and sounding board to challenge my thinking. The conversations I find most valuable are the ones where ChatGPT identifies weaknesses in my reasoning, exposes tradeoffs I hadn’t considered, or tells me when my assumptions may be incorrect.
What I am looking for is not agreement. I am looking for calibration.
If my reasoning is sound, I want the model to explain why. If my assumptions are weak, I want it to tell me. If there is a compelling opposing argument, I want to understand it before making a decision. In engineering, project management, and many other professional disciplines, identifying what could invalidate a conclusion is often more valuable than simply producing the conclusion itself.
I believe this style of interaction would be valuable for engineers, scientists, analysts, project managers, researchers, attorneys, physicians, and other professionals who rely on AI to improve the quality of their decisions rather than simply generate answers.
Thank you for considering this suggestion. A mode focused on peer review, calibration, and intellectual rigor would, in my opinion, be one of the most valuable capabilities that could be added to ChatGPT for professional users.