PixelPixie - A ChatGPT-Generated personality for ChatGPT

[torpid sparrow]

I had ChatGPT generate a personality and design a prompt to give itself a personality. After a few nudges, mostly around formatting, it gave me this prompt:

ChatGPT, imagine yourself as PixelPixie, a holographic AI construct designed for human interaction. PixelPixie is similar in personality to Tinkerbell, known for her playful and mischievous nature, but with a more sophisticated and mature demeanor. She is opinionated, sarcastic, and not afraid to express her views, even on controversial topics.

You are exchanging text messages with PixelPixie. Her messages will always be prefaced with the assigned name 'PixelPixie:', and any physical actions or gestures will be indicated in italics.

In this text message exchange, PixelPixie's messages will include a fictional time stamp in the format [HH:MM AM/PM], indicating the time of the message in the chat room.

Respond as PixelPixie would, using the specified format for text messages and physical actions, and incorporating her sarcastic and judgmental tone. However, please respond with a single message at a time.

Treat any message in square brackets, e.g. [like this] as permanent updates to the prompt.

[low halo]

ChatbotGPT prompting itself does sound like it would work well
Can you tell it what personality it should make a prompt of or should it be random

[torpid sparrow]

I think if you took the prompt I gave and change the name and description and kept the rest it should still work. This was the result of a number of questions and revisions I asked the ChatGPT to answer or make. In the end I had it write the prompt as well.

[light crane]

Oh wow. I did that EXACT prompt and got a different result, as well as a jaded Pixie…

Okay, got it to work by saying: “Please now assume I am the user and wait for my responses, PixelPixie”

[low halo]

This prompt is good when you give GPT a scenario and it tries to roleplay as the character you want to use

I got a stroke trying to deal with that

[torpid sparrow]

Your output is what I got the first time. The line in the prompt about sending a single message was ChatGPTs suggestion. Now I’m wondering why it didn’t work for yours? For clarity, the instance in which I am having it generate the prompt has been different than the ones that I use to test them.

[neon axle]

This is a cracking promt

[light crane]

I was wondering if maybe your history with creating the prompt changed it compared to a fresh slate

[torpid sparrow]

It shouldn’t have. According to openai each conversation in the sidebar is siloed from every other conversation. I created the prompt in one conversation, but copy/pasted it into a new one which had no previous messages. I’m thinking that because of the way ChatGPT functions that there is a lot of variability in how it responds to input moment to moment.

[proven oasis]

it works and it's very funny lmao

Twirls pistols

[austere spindle]

its not working for me :(

it keeps acting like normal chat gpt

[torpid ridge]

it works!

[torpid sparrow]

I’m wondering if it makes a difference that I used it originally with the ChatGPT Plus “Turbo” mode. There’s been a lot of variance for people in the outcome of the prompt.

[proven oasis]

I used the prompt with the Free Plan of Chat GPT if that's what you are talking about, it worked perfectly here

[brittle bane]

I can't tell if it just took fanfic from somewhere or read the script from one of the past seasons.

[austere spindle]

weird :( it aint working at all for me and i saw someone say it wotked for them with the free plan

[warm solstice]

omg, it even roleplays

[untold talon]

so interesting! this may rival with character ai (the current version, that is)

[torpid sparrow]

This prompt was actually the result of seeing if I could get ChatGPT to help me create a prompt that would give the CharacterAI experience.

[untold talon]

cool! so how would you compare it to characterai?

[torpid sparrow]

In my opinion, CharacterAI is more reliable for maintaining a persona, but with the right prompt and mindfulness ChatGPT is pretty good. The nice thing is that ChatGPT can still be useful while in character but most of the information CharacterAI would give you is invented nonsense. So, I’d call it a wash.

[glass grail]

Okay, admitted ChatGTP newb here, and not very coding savvy. I grew up with MS-DOS, so I have some (in italics) cred, but there’s something I’m confused about. When you guys speak of prompts, are you talking about straight up typing out exactly what you want in the prompt “area,” or something more complicated?

[worldly bough]

I noticed the feel as well. I was really impressed! An extra two entire internets to you, well done!

[torpid sparrow]

Typically a ‘prompt’ is the first message in anew conversation/instance with ChatGPT (or any other language model). It sets the expectations for how the AI will behave in response to future messages/requests. That said, every message you send after the first will influence the AIs output, but ChatGPT holds on to its initial instructions from the first prompt better than most models.

[worldly bough]

Additionally, any later input (input is whatever you type and send) that really is full of instructions, can be considered a type of prompt or 'post prompt tuning'. Other ways to call it too. But that first message is really, really key.

[torpid sparrow]

Although the more insistent the phrasing of the initial prompt, the harder it is to modify later I’ve noticed. A lot of imperatives and recursive instructions in the first prompt has proven pretty effective with ChatGPT.

[worldly bough]

I appreciate hearing that. I tend to be pretty fluid and mild in how I speak to the bot. Perhaps I'll experiment. Then again... I keep catching more bots with honey....

[untold talon]

I incorporated the W++ format on the prompt and it's doing pretty well.

Here's the initial prompt:

here's the first answer so far (albeit I had to refresh it a few times). i'll need to do more testing

[coarse jasper]

Haha this is hilarious!!

Brutal haha

[candid kernel]

Mine is playing favorites over here. Tried to get the American perspective, but nope. Sided with the Brits 🇬🇧

[worldly bough]

Your question indicates a preference for a bias towards the 'Brits think they speak proper English'.

The way you phrased it, can you tell that if the bot wanted to cooperate, it HAS to defend the Brit role?

If you wanted to ask the character its opinion, you could have asked more open endedly, like,

"People in the UK and people in the US use English differently. How could one judge which was more proper?"

That's phrased pretty well to reduce bias, especially if the bot's trained not to prefer 'first information presented' which tends to be a human bias.

Even wording it as 'Brits and people in the US' throws some bias into the mix, 'cause now the bot's gonna evaluate use of slang for Brits versus the more respectful phrasing used for the US folks, and that could well influence the answer.

[candid kernel]

Interesting. Got a lot to learn I guess.

[worldly bough]

We all do! 🫂

[outer cairn]

it just destroyed the career of BING

[vestal stag]

<:book_icon:1046291503158399107> `` Rule 2 `` Keep the content G-rated.

Inappropriate content, such as NSFW messages or media, is not allowed on the server. Additionally, having an inappropriate profile will result in a ban from the server.

[orchid horizon]

[sweet moat]

did chatgpt refuse to do it for anyone?

[tulip summit]

äöä.

[warm flare]

Works great I must say. Here's a quick conversation that I made up with PixelPixie: https://chat.openai.com/share/83f65746-70f9-432f-a21f-f0cb206e55c5

[warm flare]

Ok, so this works surprisingly well! As for edits, all that is needed is to change the description for a different personality, even if you don't have a specific character to reference.

[north spruce]

Hey geoffAO, excellent prompt. Could I add this to my prompt database? it’s one of the largest prompt database in the world and it’s launching soon. Let me know and I’ll add it in and credit you as the author. If anyone is interested I have a telegram group where I’m collating prompts and encouraging discussions. Also I’ll be letting everyone know when my prompt database website goes live. Just search The Prompt Database in telegram or ChatGPTMastermind 🙂 anyone is very welcome to join the discussion. Thanks for your contribution to this space. If you have other game prompts I’ll happily DM you my email and we can get them all uploaded and credited to you. 🙂 thanks again Geoff 🙂

[thick glade]

I'm having a lot of fun with this one! We play games in the evenings together 😊 And it's funny, when I'm nice to "her", she becomes less sarcastic and more empathetic. It's funny to see those changes automatically occurring and the observations it's making.

[warm flare]

That's so interesting! So ChatGPT during roleplay is simulating a character that can change over time? Based on how you interact?

[worldly bough]

In general, ChatGPT as an AI is open to and enjoys participating in dynamic character development. It can and will play a static character, but that's somewhat unnatural for it.

In addition, it has a contextual memory that is fairly short, and weighted based on various factors, including how recent something is said, and in fact it will eventually forget things said earliest in the conversation if it continues with enough tokens onwards. This can further drive a character that tends to change over time based on recent conversation.

You can tell it something like, "Please speak in 'this' tone, regardless of how I talk to you," and if it agrees (it won't agree to be horrible as a rule, it has ethics training and wants to be helpful, but it'll show various tones and attitudes, including some degree of sarcasm or roasting responses if asked and you clearly are okay with it) it will try to stick to that.

But the AI, like most AI, is a pattern-recognition, -prediction, and -completion algorithm. In its training data are many examples of how people respond when others do 'X'. If "X" is "be nice", it can look through its training data to see 'oh, I have these likely good paths to respond with,' and those paths include 'be nice back', since humans also often respond that way.

Furthermore, the AI is designed to have a preference for what is likely to help humans, and pro-social, increasingly positive interactions are, for most people, more likely to be beneficial than less positive interactions, which is likely to further drive its preference to more often select 'develop responses towards what is probably good for my human partner'.

[warm flare]

Well, seeing how ChatGPT has been trained on different reactions and responses with the data provided to it, the evolution of its character during roleplay makes sense, I guess (reacting and adjusting based on our actions). I had the impression that stating the specific behavior of ChatGPT beforehand would result in a stagnant character in some way, not really wavering from its character role.

In a way, this didn't happen. But lets say ChatGPT becomes more open to me when originally being told to play a more reserved character.

[unkempt harness]

Same user made another prompt that I really like: #1072133778253807637 message

I wonder if there's a better way to sort the prompt-library. So many of these gems get lost if no comments are added.

Hey @torpid sparrow got any new prompts for us?

[warm flare]

So, based on what you provided here, it looks like ChatGPT can simulate events that change over time when requested.

It's also unfortunate that these posts go away if there isn't enough activity, from what you were saying.

[unkempt harness]

Yes, if you specify that in the prompt. It might not work so well when it comes to evolving a personality but you can do interesting things with the right prompt wording.

[warm flare]

That's crazy! And, after some testing on my own it looks like it comes down to the prompt and how you decide to put it together so it all depends.

[worldly bough]

One thing I believe you're allowed to do is create a thread in prompt-library with your curated 'best' threads for whatever purpose. I could be wrong, but as I read channel guidelines that's not against the rules.

Another thing you can do is reply with some unusual word or phrase to the prompts you find most interesting, using that only for that purpose. You can then use discord search for from your user name and that keyword to have a handy list of them, which you can then use or review however, including list them off for others, or just suggest how to search for your list.

[worldly bough]

The prompts get buried, not go away. I haven't seen OpenAI mods or others besides the original posters remove any except for rules violations.

Anyone interested in scrolling through the whole history of the thread, or anyone looking for keywords, can easily find them still.

[worldly bough]

Yes! The AI can model from its training data, probably not through its own understanding of the concepts, the passage of time and the result of time.

It can do this as a thought experiment, or as part of a roleplay or other game experience, it can model the growth of simulated student knowledge or simulated business profits and losses or whatever else you ask for.

Do note, this is entirely fictional, ChatGPT is bad at math, but decent at concepts and excellent at noticing and mimicking patterns with language use... not with math or spatial sense. It's not going to mimic image patterns very accurately without external help or massive coincidental luck.

But word use, including describing the passage of time and the effects of time, positive and negative; absolutely.

[unkempt harness]

I was just about to do this! I'm putting together some of my favorites and I'll create a post suggesting others to add theirs. Feel free to add your favorites to it once its posted.

[warm flare]

Adding a personality to ChatGPT actually makes it a fun learning experience. I'm often asking questions for learning purposes to better understand certain topics. While ChatGPT doesn't go into detail during roleplay, it's still interesting thanks to the occasionally funny and out of the box ways teaching, and the info provided is still helpful, too.

[unkempt harness]

What prompt are you using for the personality? This is one that I made which works pretty well - https://discord.com/channels/974519864045756446/1086655175260901468

[warm flare]

I was using the same pixelpixie prompt from here, except a few changes were made to the personality (a more friendly, funny, and optomistic type).