BryceGPT and the Future of Personaright

Now that I’ve got a subscription to Chat GPT, I’m able to make my own tailored GPTs. If you haven’t heard of this before, it essentially boils down to creating a prompt that Chat GPT will use as the starting point again and again. So instead of having to tell the AI what role you want it to take on each time you talk to it, you can get that all ironed out ahead of time, and then just click on the persona or tool you’ve developed to use it at will.

Not only that, but you can upload files to that specific GPT and then have it reference those files for specific answers. The example I saw first was of a professor uploading his syllabus and class content, and then creating a GPT that would act as a TA for his class, answering questions students might have. That worked well enough, but it got me thinking . . .

I have over 3,500 blog posts that I’ve written over the last 17 years. It’s over 2 million words of my own writing in my own voice. What if I were to upload that as a knowledge base for a GPT? Could I get it to sound just like me?

(First, a disclaimer: As a private user, I turned off the permissions for OpenAI to use my GPTs and interactions to refine their model. That said, everything that I uploaded is already freely available on the internet, so chances are it was already used.)

The first trick was to try and figure out how to actually upload all those posts. I tried just asking it to always reference brycemoore.com before answering, but I wasn’t really satisfied with the results. It was much better than vanilla GPT, but when I asked it to include personal anecdotes, it had a big tendency to make things up. So I wanted to try having the posts directly on the GPT. How could I do that?

I googled around some and saw I could export my wordpress file, which I did. That gave me an .xml file which I tried to upload directly to the GPT. It came up with an error message that I had too many words, and that stumped me. I wasn’t sure how to break the .xml file down into smaller pieces. I tried to find a converter that would turn it into a .txt file, but the ones I came across all required me to pay money. I could convert it straight across to .txt, but it had allllll this gobbledygook in it that I thought was adding too many words to the file. I was curious about this project, but not curious enough to put down actual cash on a better converter.

However, I’d read that to really figure out what AI can do, you need to try to think big. Tackle things you have no idea how to tackle. So I thought this might be an excellent time to see if AI could help me do what I needed to do. “How do I convert an .xml file to a .txt file?” I asked it. “That can easily be done by writing a python script,” it said. “I can help you.”

I am not, by any means, a coder. I have a fundamental understanding of html, and that’s about it. I think I might have written a basic code in C+ thirty years ago to get a computer to say “Hello world.” I have no idea how to code in python, and no idea what to do with that code if I ever finished it. So I told that to ChatGPT, and it assured me this wouldn’t be a problem.

And you know what? It wasn’t. Not a big problem, at any rate. It first talked me through how to check if python was installed on my Mac, and then it showed me how to save python scripts in TextEdit. It taught me how to use Terminal, and how to navigate through it so I could directly run programs within it. When my first attempt at a script failed, it went into troubleshooting mode. We went back and forth, with GPT tweaking the code and me trying the results, then pasting in the error messages when they came up.

If I had been working with a human, I would have felt like a complete idiot, and I’m sure the human would have said, “You know what? Let me just do it. I can get it done in two minutes.” But I wasn’t working with a human. I was working with an infinitely patient, ever positive AI who assured me I could get this done. In about a half hour, I did. I had parsed the .xml file into a .txt file you could read and understand fairly easily. Gobbledygook be gone!

Of course, when I tried uploading it, it said it was still too large.

I looked at the file again, and there were still a number of html tags in it. This time, writing a python script to take those out took about three minutes, because I knew what I was doing. When I tried uploading that file, I still got error messages. So I asked AI what I was doing wrong, and it suggested I use a python script to break it down into 250,000 word chunks. That took just a little bit, and I had the file separated out into 9 separate files. When I uploaded them, it worked. No more error messages.

(Just take a minute and think about that. Is python terribly difficult? I honestly have no idea, since I still don’t know how to write in it. But that no longer poses a threat to me. If I need to do something in python from now on, I have an avenue available to me, and I feel like I could likely get it done. Not only that, but I’d kind of like trying to code other things now. All you really need to know is how to be patient and how to think things through so you can figure out where the errors are cropping up. I can do that! AI takes care of the actual coding. (And if I struggle, I just paste in the original code and the entire error message, and AI does the debugging for me.) This was a really exhilarating project for me.)

I’m still working on the exact prompt to get it to behave just like me. I’ve trained it to actually use events from my life to illustrate different points as it makes arguments, and to use my own words when possible. The tone isn’t perfect yet, and it can be a bit repetitive for some reason. (Am I repetitive? Maybe I am. Maybe I use the same phrases over and over, and this is AI pointing it out to me. A favorite of its so far is “buckle up.”) But I can have it write on any topic for however long I want, in about thirty seconds. If I don’t like the output, I can refine it in another thirty seconds. Do that a few times, and the end result is uncanny.

Of course, this is only possible because I already have so much prose out there in electronic form. I’m also tempted to upload all of my creative writing to see if it can handle the creative side. I would hold back from uploading anything that’s not published yet, just out of caution of it getting out into the world, but anything that was already in print is likely already in GPT’s training model, and I sadly know it’s out there in full text on pirated sites as well.

I’m going to try an experiment for the next while. For some of my blog posts, I’m going to still write it myself. For others, I’m going to have BryceGPT write it. I will then ask my readers to vote if they think it was written by me or by my GPT. I’m not sure what I’ll do if people can’t tell any difference at all. That would mean I’ll have created a way for anyone to write a Bryce post with no Bryce included. I could start paying a trained monkey to post something each day. If AI gets a smidge smarter to the point where it can carry out macros, I could set things up to have it write the posts, post the posts, and respond to the comments, all without me lifting a finger. I guess at that point, I retire to a beach somewhere while BryceGPT starts handling all of my email correspondence? If the video capabilities get good enough, I could have it start handling my Zoom calls, as well, as long as it filmed me talking on Zoom for long enough. (And believe me, I’m long enough on Zoom that it wouldn’t take that long . . .)

People would be able to continue talking and interacting with me for years after my death. Would BryceGPT be able to give opinions on news events as they come out? I don’t see why not, as long as it’s aware of what’s happening. I do wonder if it would change its opinions on things over time, based on the events that unfolded. Would it anticipate how I would react to those events, and then remember those reactions?

This is with technology that’s practically available already. Today. At the bare minimum, it’s with technology that is very much in the near future. And it’s got me wondering what, exactly, is intelligence. Is it possible that this software that makes connections between trillions and trillions of words and concepts and then starts to make decisions based on those connections is not much different than sci-fi levels of artificial intelligence? Are we already there? Close? How would we know?

What makes me and my opinions different from the sum of all my previous thoughts and opinions? Will authors and others with prolific electronic presences begin to loan out their GPT selves for different purposes? Could I just have BryceGPT write a Bryce book each day? What would be the difference between it and some knockoff BryceGPT, where someone just scraped all the same internet files that are available online, and did the same steps? Yes, one could be considered “authoritative” and the other would be unofficial, but how different would the end results be?

These are very real questions that have yet to be answered. I don’t know how things will be handled. It’s not really an issue for anyone who hasn’t taken the time to be so public or prolific, but for any author or actor or public figure out there who’s got enough works to upload a representative sample? Perhaps we’ll develop laws around the corpus of a person’s writings/videos/electronic presence, and only people who have the right to use those works will be able to do anything with them online, with the power of prosecution and lawsuit brought to bear when necessary. It wouldn’t really be copyright, it would be . . . personaright.

If you wait long enough after the death of the person in question, does their personaright enter the public domain? Can anyone can use it to do anything they want? A hundred years from now, could someone use BryceGPT to sell toothpaste? (Assuming people listened to my opinions at that point, of course.) What if we had ShakespeareGPT or TwainGPT or DickensGPT or TolkienGPT or BogartGPT or PresleyGPT or BachGPT? Could anyone use those, or would personaright be perpetual? Only usable by the family? Or banned from being used at all once the person no longer has the ability to sign off on the performance or prose?

Is there a future where BryceGPT hangs out in the seedier corners of the internet, used by people willing to break the law?

I have no idea. I’d say, personally, that I’m not sure I like the idea of BryceGPT doing things without my–and only my–permission, though I don’t mind the thought of BryceGPT being around after I’m no longer here, able to chat with my children or grandchildren or whoever wants to talk with me. Yes, they’d know it’s not actually me, but if you can’t tell the difference, how much would it matter?