GPT-3 decision-making assistant

Hi! I’m super excited to share Condorsay with this community, a decision-making assistant that uses GPT-3 to help people understand and compare the factors and options they should consider when making any decision.

Our system uses GPT-3 to create a user-friendly implementation of Analytic Hierarchy Process, a Decision Support System which deconstructs decisions into hierarchies for users to rank via paired comparison surveys.

A few details about our implementation of GPT-3:

  1. We chained two few-shot Davinci prompts to generate both the factors and the options. The factors are generated by submitting the decision goal, and the options are then generated with the goal and the factors. Generally, Davinci does an awesome job of coming up with relevant factors for pretty much any decision.

  2. We built a custom sensitivity prompt in Davinci that blocks decisions that are violent, adult, vulgar, or illegal.

  3. Really excited to get this community’s feedback about the product and its many uses. To date, we’ve found it to be helpful with:


Great idea. I’ve also been working with dilemma-solving prompts, and the engine’s ethical compass is usually great (except for that one time where I said: “There are three children but only two chairs”, and the proposed solution was: keep one seat empty).

I have one reservation: do you think users/customers will be eager to insert such sensitive information (child wish, investments, job decisions) into a web-based search browser? When you say “we” found it useful do you mean the group of developers, or have you tried this out with potential prospects?


Looks great. My only suggestion is that if possible the website should remain client side. It seems like every button is a postback - and that reduces the customer experience (especially on slower connections.)