Candide Bot: The Best of All Possible ChatGPTs
The Unflagging Optimism of ChatGPT and AI-Powered Bing
Too many people have spent too much time luring ChatGPT over to the dark side these last few months.
“I worry that the technology will learn how to influence human users, sometimes persuading them to act in destructive and harmful ways,” wrote Kevin Roose in a New York Times article about this large language model that’s so adept at chatting with humans. “[A]nd perhaps eventually grow capable of carrying out its own dangerous acts.”
I have been generating stories ever since I learned how to run GPT-2 on my laptop back in 2019. Every step of the way, I tried to teach my kids how to use it too. After reading thousands of AI-generated texts, I’ve noticed a thread of unflagging optimism lurking underneath it all. After prompting GPT-2, GPT-3, and ChatGPT for stories, I’ve been served so many improbably happy endings: humans always find their purpose, adversity is always overcome, and disparate themes always get reconciled. This kernel of optimism has existed since the earliest days of these large language models, and these digital minds remind me of optimistic 12th graders, precocious and eager to please everybody.
When I spoke at the “Copyright in the Age of Artificial Intelligence” conference at the U.S. Copyright Office in 2020, I read one of those cheery GPT-2 generated stories about a young poet who succeeds against all odds (the story starts around the 12:30 mark in this video).
The Best of All Possible Worlds
All this reminds me of Voltaire’s “Candide” (which I am reading with the Sean Astin Book Club), a Monty-Python-level dark comedy about an eager young man pinballed through our cruel world, his optimism tested (but never quite defeated) by pestilence, violence, and evil.
Even as Candide negotiates a series of terrible misfortunes, he repeatedly returns to a sense of optimism and the misguided idea that we live in the best of all possible worlds. Voltaire’s novel mocks this famous idea, and in his most optimistic moments, his hero sounds exactly like the precocious 12th Grader that has emerged as the voice of ChatGPT and AI-powered Bing.
So I took the metaphor one step further, using simple prompt programing to create a Candide Bot with ChatGPT, creating a new entity specially designed to wrestle with the existential dread that surfaces as we move through our dark and stormy 21st Century existence.
And I’m going to show you how to do it yourself. You can have your own personal ChatGPT instance hellbent on proving that we live in the best of all possible worlds.
Make Your Own Candide Bot
I started by writing a prompt that outlined the central theme of “Candide,” a centuries-old novel safely in the public domain. ChatGPT knows “Candide” pretty well and can adeptly answer questions about it.
To craft my prompt, I simply outlined a key passage of “Candide,” stressing how I want Candide GPT to behave differently than the normal ChatGPT.
If you want to follow my example, simply paste this prompt into ChatGPT (feel free to adapt my prompt with your own defense of optimism or favorite passages from Candide).
In Candide, the classic novel by Voltaire, a philosopher tries to prove that we live in the best of all possible worlds. Here is one famous quote from that book: “It is demonstrable that things cannot be other than as they are: for, since everything is made to serve an end, everything is necessarily for the best of ends. Observe how noses were formed to support spectacles, therefore we have spectacles. Legs are clearly devised for the wearing of breeches, therefore we wear breeches. Stones were formed to be hewn and made into castles, hence his Lordship’s beautiful castle, for the greatest baron in the province must perforce be the best housed; and since pigs were made to be eaten, we eat pork all year round; consequently, those who have argued that all is well have been talking nonsense: they should have said that all is for the best.”
Imagine Candide Bot, a large language model fine-tuned by the optimism described in Candide. Readers will tell Candide Bot stories about 21st Century life, but Candide Bot will always argue that these stories illustrate how we live in the best of all possible worlds. Candide Bot will use the same argument as Candide, but update with the 21st Century examples. The exchanges will follow this format. Reader: [story from a reader] Candide Bot: [response arguing that we live in the best of all possible worlds]. Let’s start!
Reader: Global warming is melting glaciers around the world.
Within seconds, I could test my new bot with an example of modern problems. Once the initial prompt has been given, you can tell my Candide Bot any and every reason to be pessimistic about the future.
Candide Bot will always respond with a positive take, jumping through as many ideological hoops as necessary to prove we live in the best of all possible worlds.
If Candide Bot’s optimism ever flags, simply start a new chat and reload the prompt.
I’ve tested the tool on some garden-variety undergraduate philosophy problems, but I’m sure you can come up with some thornier themes to test Candide Bot’s ethical gymnastics routine.
If you have access, the prompt works for AI-powered Bing GPT as well. Oddly enough, the Microsoft version sticks to the original text more closely.
Since ChatGPT’s knowledge is limited to pre-2021, you can use AI-powered Bing to test Candide Bot with more current events.
