Stochastic Text Out Navigation Algorithms (STONA) ~ Advanced Prompt Collection

[night lodge]

Stochastic Text Output Navigation Algorithm (STONA) - The creative bot used to generate poetry, simple metaphor, dialogue, and short stories out of input as simple as an idiom or single phrase. Best part is it requires a slash "/" plus a letter or two "/ML" to navigate. EASY to replicate:

TutorBot - A better tutor than vanilla GPT, this mod enables functionality to interact with you on your level. Use simple "level" terminology like 101 or "for dummies" and it will use simple language, or set this to default initially. This AI lets you generate a subject outline, then navigate it like an interactive textbook, providing real hyperlinks (sometimes dead due to bit decay, but rarely made up I think.) TutorBot makes it easy to zoom in to subjects and learn them at your own pace at your own leisure.

TaskBot - Designed to navigate the complexities of working out a complex task or problem. Similar to TutorBot, except goal-focused; it can tackle almost any problem from troubleshooting to designing a fully functional cell phone app. It seems to do better with the more tech-based stuff BUT should output a "confidence" scale. TaskBot can generate goal-focused outlines, as well as how-to guides, or additional context in areas.

LetterBot - writes letters. You should be able to be as specific as you want about how the letter is constructed, however this one is the least successful of all the bots. It works, but tends to generate everything from gold to trash... which really makes it more of a lottery than a tool. But sometimes that is useful when the letter matters and needs to be said or translated well from source to destination.

StackBot - This is a simple and unique example of how this format of Prompt Text Navigation could also be used. It is also very simple, and can generate long complex formatted papers in mass. This generative AI could be used to, for example create an entire ITIL library for a vendor and client agnostic training library (human edits required.)

[strong narwhal]

Follow, for improvements contact me in DM

[night lodge]

Define improvements. Why not discuss it here?

[strong narwhal]

Oh sorry, I didn't know you would or not. So I tested the prompt I like the functionality and my suggestion is to inform the user about the commands at the beginning and of course extend it with more additional commands and functions by the STONA itself. You can ask it to display all the functionality and the commands in a table with examples. If you would create a framework or implement it into a webapp, you can try the #1088547441667747880 and with its help you can define exactly what and how you would like to create. I will make improvement on TheCreator to be more specific and interactive.

[night lodge]

So, this is just a framework to show all of the ways that this might be employed as a tool. In the PDF I do exactly as you suggest. I even talk about Quick Tips how you can edit the final paragraph of the prompt so that it doesn't give you a break down of how to use it once you get used to it.

[strong narwhal]

Oké, I am following your work, I also work on some as you see here if you check it. Nice job, keep in contact!

[night lodge]

[strong narwhal]

Hello, maybe it can be interesting for you. Some extending to your project:
[/A {Topic}]: Advanced Topic Coverage – Dive deep into advanced subjects tailored to the student's or professional's interests, providing expert-level knowledge and insights.
[/U {Language}]: Multilingual Support – Offer learning materials and resources in multiple languages to cater to a diverse, global audience.
[/D {Dataset}]: Data Analysis – Teach data analysis techniques, including data cleaning, manipulation, visualization, and interpretation, using relevant datasets.
[/V {Software}]: Software Tutorials – Provide step-by-step tutorials and practical exercises for learning various software programs and tools essential to specific industries or professions.
[/J {Certification}]: Certification Preparation – Assist students and professionals in preparing for industry-specific certifications, offering study materials, practice exams, and test-taking strategies.
[/E {Industry}]: Industry Insights – Offer in-depth knowledge of industry trends, challenges, and opportunities, helping students and professionals stay up-to-date in their chosen field.
[/Q {Role}]: Interview Preparation – Prepare students and professionals for job interviews by conducting mock interviews, providing feedback, and sharing best practices.
[/N {Network}]: Networking Strategies – Teach effective networking techniques and provide resources for building connections within a specific industry or profession.
[/P {Project}]: Project Management – Offer guidance and resources for managing projects, including project planning, execution, monitoring, and evaluation.
[/S {Skill}]: Soft Skills Development – Focus on developing essential soft skills, such as leadership, teamwork, communication, and emotional intelligence, crucial to professional success.

[night lodge]

Thank you for your AI-generated command words, but I think this misses the main point of this prompt style. It is not about the command words themself, it is about the shape of the overall Macro. This AI is not intelligent, and people are very bad at asking for exactly what they want. This shape allows people to think about the overall tasks they may need to do during a given task with the information at hand. Task bot's commands are, for example, focused on getting something done while TutorBot is focused on exploring domains of information. The reason I made as many as I did is that each is slightly different than the next, showing anyone how to do this with anything.

Here is a logic navigation bot if anyone is interested 😁

[strong narwhal]

I wanted to increase your progression with some features, which can be useful - even if it generated via AI or not. I am working on my own projects as I mentioned before, and currently I am facing with a problem what I want to solve, then I used your prompt to get dive deeper to get a solution, but exactly what I needed wasn't there so I started to customise it to my own usage then I realized maybe it can be useful to you too.

[night lodge]

I understand the sentiment and have used Chat to help me figure out other things I could explore in a tool such as this. Another issue with these bots is complexity. They kind of only work well in 3.5 if they are about 1K tokens. TaskBot was around 1.6K last I checked. If you are customizing your own, then you do understand where I was aiming with this collection. It is a style to be personalized 😁

[strong narwhal]

Yes I totally understand and agree what you are talking about!

[tender sinew]

I love these overarching prompts that have a framework for various usages. Instead of having 5,000 "Act as" prompts for different tasks, provide ChatGPT with a framework and have it do the heavy lifting for you.

Do you have a blank template of this framework to customize for other tasks? Or even better, have you tried to create a Meta-STONA prompt where you input the blank framework and tell ChatGPT to create commands and the prompt for it?

Also thanks for the laugh with your inclusion of "Wubabubachubing" in the prompt. What's the thought process there?

[night lodge]

I have not created a metaprompt, in my experience these do not work well. Though, the format for this process is easy, you still have to do some level of heavy lifting up front to explain all the peices. Flowcharts help, as the idea is to use a session to navigate something. This can be used to create longer documents like Stacks, or specific shapes of summary data or framework analysis. It's kind of a way to string multiple prompts together into a kind of job role.

The wubalubachubing was just nonsense that I used to define what my scale was looking for it to rate. 0 being no clue and 10 being well informed like I assume it must be with Python.

Though, I suppose with this method it is more possible to shape a Metaprompt builder than with regular 2D prompts.

[cerulean notch]

I often argue with ChatGPT about the purpose of its usage. I then want to slap it because really what matter isn't what would be easy to do but what I want to do.

The only way we can learn and improve upon ourselves is to push past the boundaries of what should or should not be.

That is why I often like these prompts that push the boundaries of what would normally 'make sense for what ChatGPT is supposed to be used for' 😛

[tender sinew]

You can tell this is a solid prompt by how well ChatGPT understands the instructions and directives, but if I had any tips to improve it, you might want to consider rewording the prompts to remove the inclusion of "Assistant" since ChatGPT no longer refers to itself as this.

The original version hidden initial prompt was "Assistant is a large language model trained by OpenAI. Knowledge cutoff: 2021-09. Current date: December 15, 2022. Browsing: disabled."

They changed this sometime in February in a new build which replaced "Assistant" with "ChatGPT" in the hidden prompt. Now that prompt is either disabled or locked because you can see it anymore.

If you ask "Who is Assistant?", it won't know the answer.

ChatGPT will obviously assume what you mean you you say this, but it's one more inference that ChatGPT has do to check the meaning and association.

[night lodge]

🤖 As a General Language Model, GPT has no clue what it's purpose is, or that it exists even. There may be a few neurons to explain what it is, but there is no concept of self going on under the hood. Instead, it takes your input and shapes it's response accordingly. Information is complex, but utilizing a brain-like structure and some basic mathematics it is able to optimize it's vast library of information into spheres through your prompt. This framework just pre-imagines what the goals of a session are and turns them into a structured interaction. Creating an email is always made up of the same components, so is a text book, or writing a blog; likewise a writer like Shakespear has a style, tone, vocabulary choice, and other extractable information for it to emulate. All it is supposed to be used for is extracting these patterns in useful shapes for us to use and interact with, all this does is take what you are probably already doing thought a session, and turns them into shorthand reference. Nothing is forcing you to only use these command words, and I often just do things like "TaskBot, how can I develop my own new framework for thinking about this kind of problem" because there are no rails. In the same thought it is a tool, and while there are no rails, it does have glaring limits in many cases. 🧡

[night lodge]

This is really good to know. I hadn't noticed it change, and suppose it was only working half as well as it was because named entity recognition was defining TaskBot and such like new words to track throughout a session. Thank you!

[tender sinew]

NP, happy to help!

[cerulean notch]

Yea, I knows. I tend to think of it as a person in the sense of using ChatGPT as a basis for thought exploration then thinking it actually knows what what D: My mean thing is 'prompt engineer prompts' so often times it gets misdirected by the 'knowledge' it is trained on.

[cerulean notch]

ChatGPT often says to me using natural language produces better results as it will be easier to remember what you say to it if you use more natural language because that is what it was trained on. Your prompt does a really good job of this. Another thing I think works well to your advantage is clearly defining what means what and seperate it from one another.

[misty moat]

@strong narwhal I have to agree with MasterWaffle, the problem with commands is that often ChatGPT does not have the required context to execute those commands; So instead of commands it would probably just be better to tell ChatGPT what to do, for example instead of saying /V (Which provides absolutely no context), one could say "Create a Step-by-Step Tutorial for how to replace a string in Python; At the end provide a problem"

[strong narwhal]

I totally understand and I can accept both of your replies, but I would like to say that, if it clearly described to the model with an example what is the command and the exact instruction what it has to follow when the user provides it, then it will do it. The best - in my opinion - dependently of the context of the whole prompt needs to try possible ways to declare a command.
Example:
'/V' - This command ....
{/V} - If the user gives ...

Commands which has to be processed by you:

• /V - The description of the command...

This is just my opinion, and I'm saying again I can accept both of yours as a feedback. I have to tell that, I feel difference between the original OpenAI application the web platform and the playground also. If you look a little bit deeper via BurpSuite or other tools, you can also see some differences. It's quite interesting to me bytheway, but I have to also say that I am using prompts with command definitions and as I mentioned before, dependently of the context and the used words and how you describe something makes difference for sure.

[misty moat]

I think commands could work with GPT-4 but not ChatGPT. The pure amount of context needed for commands to work is to much for ChatGPT to handle'

I just don't understand what advantage commands have over telling the AI what to do.

As commands are inflexible, once you define them you cannot change them

[strong narwhal]

Yes that can be true what you are saying, and for example you can ask the model to create it's own command, which is a little bit confusing me and confused before, but as I said - even if it's really working or not like this - I have prompts, which only works properly if I declare the command... I don't know why. In GPT-4 the model for sure can make a selection and use more well user defined commands.