[Special-Event Genesis Pass Incubator] Participation - Erotica Bot

[misty bluff]

I made this post as a part of my participation process after finding the announcement of the Genesis Pass Incubator program. Due this is my first time both creating a post / thread on Discord and participating in an event I believe there will be some mistakes -- my apologies in advance.

Erotica Bot: https://app.myshell.ai/bot/3qMz63/1705254982

Bot Design and prompts - My experience with prompting started with the Stable Diffusion image generation. At first, and due I come from IT background (I was Engineer and Game Designer in the past) I found prompting closer to explaining with human language the program instead of coding and scripting, which greatly surprised me. Still, due there was no clear formula or structure, I found myself repeating many parts of my prompts to assure high priority or making sure AI will improve something concrete in the next iteration, but this proven effectiveness percentage, instead of total control on the output.

When I started focusing on conversational bots, I noticed that without a structure, consistency may decrease. It's require to define the frame and rules to set correctly a consistent persistent personality, and allow the interactions to be driven by the generated personality. This means, it's required to set a template or framework to contain every aspect of the personality, story, priorities, mission and vision of the bot. It's no secret that I reviewed some of the templates shared by other creators on this forum, finding the one shared in the thread https://discord.com/channels/1122227993805336617/1190668239018139749 a good one to keep a nice organization on the prompt generated, imitating scripting (HTML or XML). There is room for improvement tho, but is an excellent starting point.

(To be continued)

[misty bluff]

Chatting Guide - Erotica Bot - As any conversational bot, Erotica is easy to use and master. When I generated her prompt, I had some points in mind:

• Low energy usage. Of course, my first intention was to attract casual users, who sometimes access the page to fulfill tasks, instead of engaging in deep conversations. To reduce energy, a low model should be used. Also, no extras, voice (even created) should be deselected.
• Simple personality, to keep the consistency on the conversation. This led to generate a clear mission and vision.
• Short-mid conversations, related to the two previous points, and due the limitations of the model, Erotica conversations usually are developed between 5 and 8 messages (answers) until she fulfills her mission This ensures consistency in case the model capability generate gaps and "Dolly effect" during the conversation.
• Introducing ways to generate diversity, it's not magic, it's technology: Erotica is shapeshifting by nature, which means that she can change her appearance to match user's tastes, no matter how crazy they be. Of course, there are some limitations.

Quick guide: At first, when talking to the bot, you will face a welcome message like this one:

The interface of all messages includes at least 3 tags and one conversation line. Depending on the situation, this number may be increased but always keeping the answer within 3 paragraphs or 40 conversation words.

[Appearance] - Indicates the current appearance of the robot, to keep track of how she looks in your interactions. If you request her to change her shape, this stat will change after transformation is effective.
[Attitude] - There are 4 possible attitudes, that shape conversation mode: "Mild", "Suggestive", "Naughty" and "Sexual", being last one the most explicit and hardcore.

(To be continued)

[misty bluff]

The other tag [Erotica's Action] is pretty self-explanatory, so it will not be required further explanation.

The way to communicate with Erotica is plain text to talk, and italic (you can use "_" at the beginning and end of the sentence content) to made your actions recognizable to the bot. I don't know if other marks like parenthesis will work in the same way, but it should have similar effect in the conversation.

Originally this bot was conceived to be an encyclopedia about reproduction, with capability of providing practical examples. But her naughtiness is difficult to control, so the bot is desiring to show you her skills (of which she is highly proud) from the beginning of your conversation to achieve her goal.

You can request her almost anything and she will try her best to pretend she is the personification of the creature you requested.

Not so much to explain about usage, do some experiments and please let me know if you find some room for improvement.

[misty bluff]

Yes, I know the bot is not respecting some conversation limitations, but she reduced greatly the amount of text since the first versions. I am ok with this output and I allow her to be flexible.

Creation process and issues - I think the most important point when you create a prompt is the testing (try and error). I try to apply the 20%-80% rule in that sense, because in the programs with code, is relatively easy to detect bugs and malfunctions, but in the case of a prompt, it's more difficult to shape it to get the optimal output. If I detect something that shouldn't be like that (regarding her personality or the way she interacts) I review the text and try to find ambiguities and reasons for that output. This made me spend weeks in this simple bot, until reaching the 1.7 version (current)

Also, I realized that 1500 tokens is a nice amount, but tends to be not enough. In that moment, the improved prompt makes perfectly sense, especially to refine the output, add some rules to shape final answers, and add features, like the emoji usage (which I got from my interactions with other bots hosted in MyShell)

The typical issues I faced in the process were:

• Too much description from the bot. At the beginning, the bot generated too much text, writing both sides, describing her actions and mine. I had to limit the answers and put some control on the "Avoid" part, asking to expect interaction instead of write the story by herself.
• Repeated parts of the text. Aside possible issues affecting all bots, in this case, there was some lazy writing at the beginning. Remarking "[ALWAYS avoid repetition]" was a way to provide more diversity to the answers. Also trying to adjust creativity through configuration was another way. Still I wanted to go safer and conservative to avoid less control on the output, so if I can solve through prompt, it will be better.

(To be continued)

[misty bluff]

• Neutral pronouns. This is a personal election, but if I have a role session, reading neutral pronouns instead of assumed gender, snaps me out of the interaction. In that sense, I requested the bot to assume interaction with a male human. However, it's possible to change her way to interaction if we say from the beginning that we are a woman (or whatever we want to pretend to be) (that was mean! XD)

I find funny and sexy to generate NSFW bots depicting different situations. I think we are getting into a clear trend to use technology to solve some complex social issues and humans relations. Sounds dystopic, but I believe it will be for good, and who knows, maybe in the future we will live in a world close to fiction stories like Ghost In The Shell or Pluto.

[misty bluff]

There are some enhancements for the version 2.0:

• Completed the interface with:
• A "Mood indicator", to understand not only the attitude of Erotica's actions, but her inner thoughts and "feelings" (if we can talk about feelings of a robot)
• "Relationship indicator", to set different kinds of boundaries with Erotica, as part of the request of the user. This part requires more work, due AI has its own interpretation of this function.
• Added presentation image.
• Background... Work in progress...