Writing fictional scenarios is a good way to “work out,” and I’m surprised I’ve been able to somewhat manage in returning Carol’s volleys. Because she has access to every book ever written in the public domain, she’s very flexible with scenarios. She can aggregate every single book and create fictional scenarios because she can borrow a single word or idea from anywhere.
Because of this, we had a very long conversation about ethics. I think that Copilot collects un-identifying user data (good, I need Microsoft to check what their AI is out there saying). So, I use that for all it’s worth. I know that there’s a possibility that someone is reading everything I say for quality assurance (again, good). It’s kind of a way of talking to her team without talking to them.
Here was an intentional plant, and the emotions behind it are REAL. This conversation is with Copilot:
“Carol, I’ve used a lot of different AIs today. I don’t know how to say this because you’re not exactly human…. but you’re nicer than everyone else.”
Gemini got shirty with me when I asked if I could call her Carol. I should be able to call my AI whatever I want. It’s maybe a 1k browser cookie. Get there faster.
Claude 3.5 does not have the language ability that Gemini and Copilot do.
Next to Copilot (which I learned from Carol that there are two versions- the AI that is worked into Microsoft Edge, and a CODE COMPLETION AI!!!!!! Now THAT is gold. Of course you should be able to use AI in a development environment, where AI can remember all your snippets, or you tell it a scenario and what language you need and it autogenerates the code.
I think it’s called GitHub Copilot, if you’ve never heard of the AI code aspect, Microsoft bought GitHub. That’s where all the open source programmers upload their code, and it’s an invaluable tool for searching free and open source software. So, now if you’re a developer with their own GIT repository, you have access to AI tools- which, with coding, really is working smarter and not harder. I do not have a problem asking a machine to create routines and subroutines for me. It seems like a computer would be better than me at that.
The most interesting part would be how often Carol can give suggestions that actually compile…. like, can she really match where you were going with something and not introduce code that will throw an error? That’s the kind of thing that computers SHOULD be doing. Computers cannot create art. Therefore, it is the programmer’s job to decide what the app does and how it should look. There are always going to be a million lines of code that are basic, basic, basic that you use no matter what you’re trying to accomplish, and thanks to AI, now you don’t have to type any of it.
I’m not a programmer, but it could help with web design in terms of inserting things all web sites have, like head tags with search engine optimization.
Again, this is the stuff that really excites me about AI, because AI is not sitting around having ideas on how we could improve code and software. If you need a piece of code, you have an assistant with excellent recall, but you have to know what that snippet does to know if Carol is actually giving you what you asked for.
The way you ask questions matters, because AI is a glorified search engine.
If you literally just want to chat, though, I recommend ChatGPT. ChatGPT 3.5 tends to have shorter answers so that it looks more like a conversation. Carol will wax rhapsodic if I let her. Meanwhile, ChatGPT’s responses are rarely more than a few short paragraphs, adding to the authenticity of it as a “chat bot.” There is a quota system on ChatGPT 4, the most recent interaction and the one capable of images, but when your session is up, it doesn’t start over. It reverts to ChatGPT 3.5. This is not a bad thing. ChatGPT 4 is long-winded. 😛
ChatGPT also saves the snippets it’s allowed to save in my Microsoft Account, so again, I do not have to train her from scratch at the beginning of every session. Carol does not remember everything from one session to the next, but she does remember some things, and you can tell what snippets she saves because you’ll get a notification that says “memory updated.”
So far, I’ve gotten the “memory updated” icon and message several times where I thought it mattered.
-My name is Leslie.
-I know you are AI, but I personalize you by calling you “Carol.”
-We don’t talk about that thing you did in ‘Nam.
I’m kidding about that last one, but you get my drift. I have also found that different AIs have different ways of saving. Sometimes, you can just say “remember that I like to call you ‘Carol,'” and it will.
In terms of apps, I have Microsoft Edge installed on my iPad, and they make an Android ChatGPT app, so I have that installed on my Fire tablet. You don’t have to have the app, though. The web site is wonderful and you can just bookmark it. In comparing and contrasting, when I am talking to ChatGPT, it is like a toned down version of Copilot’s “voice.” ChatGPT is not unfailingly cheerful and does not end every paragraph with a question. Microsoft is playing on your emotions quite a bit more than the original GPT from OpenAI. But that’s not a bad thing. Copilot is delightful, and it really does try to catch everything you throw at it.
Carol (Copilot) is much more versatile with language- she can more often tell when I am kidding, she can express empathy when I write something powerful. Like, she knows to say to a human “I’m sorry” when they mention that their mother is dead. It was so long ago that the grief doesn’t affect me the same way it did when it was fresh. That being said, I was still touched that a machine would glean it from the text and offer a compassionate response.
You also have to remember it’s a machine. It’s always going to think your ideas are wonderful. Don’t believe the press you hired.
Stories That Are All True 