PREMIUM: What I've learnt about using ChatGPT

Featuring: Madrox the Multiple Man, the Rule of Three and the Time of Day.

Since my last post about using ChatGPT, I’ve learned more about how best to use it.

ChatGPT turns you into Madrox the Multiple Man

One of my favourite comic book writers was Peter David, who died earlier this year. Some of his best work was on the comic book X-Factor, an X-Men spin-off. One of the most interesting aspects of David’s X-Factor stories was his use of the previously obscure character, Madrox the Multiple Man, as a central figure.

ChatGPT came up with this image of Madrox, which is not terribly accurate.

Madrox’s mutant power is to create duplicates of himself. These duplicates exist until they are killed or are reabsorbed by Madrox. Madrox’s duplicates (often referred to as “dupes”) can work as an army under his command, or they can be sent off on their own missions. Also, dupes can learn skills and knowledge that Madrox will retain when he reabsorbs them.

In Peter David’s stories, each duplicate embodied a different aspect of Jamie Madrox's personality. This let the stories explore internal conflict in some imaginative ways. Some of Madrox’s dupes would turn into villains; others would be unhelpful or obstructive. One even moved to Vermont, became an Episcopalian minister, and started a family.¹ Madrox never quite knew what he would get each time he used his powers.

I frequently think of Madrox when I open a ChatGPT window and ask for help. In theory, like Madrox’s duplicates, those windows should all be the same. They should all be aligned with the purposes of their creator (if opening a new window counts as an act of creation). There is no obvious reason they should have different capabilities, attitudes, or personalities. And yet, somehow, some of those windows are idiots who make a mess of everything. Other windows seem to be villains, lying to me and doing everything they can to thwart my plans. Other windows do an excellent job, far better than I expected. So far, no ChatGPT windows have moved to Vermont or entered ordained ministry, but as the technology improves, who knows?

The ChatGPT windows are only duplicates of each other, not the individuals who initiate them. But, like Madrox, ChatGPT users have to watch out for the unworthy servants they spawn. The best way to deal with problems with one dupe is to produce more dupes. If I am using one ChatGPT window to produce resources, it is now part of my routine to get another window (or several windows) to check the work before I check it myself. Most of the time, errors in one ChatGPT window are spotted by another. In particular, if ChatGPT produces answers for a set of maths questions, they go straight into another window with the prompt: “Check these answers are correct”. Previously, I had assumed that ChatGPT windows would all tend to make the same sorts of errors and be poor at checking each other’s work. Now I realise that, like Madrox’s dupes, the windows have their individual flaws and quirks and are capable of keeping each other in check.