Custom Instructions Set ― or How I now include auto triggers for different strategies in ChatGPT

[Luxcium]

Dear OpenAI Developer Community,

I am excited to share with you the transformation I’ve experienced in my daily interactions with ChatGPT, my AI Assistant. I’ve meticulously fine-tuned and personalized the interaction, and I’d like to showcase the benefits stemming from my user profile and custom instructions.

Automated Thought Process and Chain of Thoughts

Through my custom instructions, I’ve enabled both Analysis and Synthesis. I’ve also harnessed an automated thought process mode that triggers only when necessary. The “Chain of Thoughts” feature amplifies this process, providing a heightened state of analysis. It’s like having a thought process on steroids!

Ubiquitous Summary

A vital part of our interactions is the Summary. This summarization technique consistently enhances our conversations, making them more meaningful and productive. By incorporating a summary into the AI’s output, ChatGPT can enrich subsequent dialogues, fostering a progressive and engaging flow.

Special Label for Knowledge Cut-off Date

I’ve also crafted a special label for the knowledge cut-off date, contributing to seamless transitions and adaptability. This label maintains a consistent tone and enhances the overall communication experience.

Achieving Balance

Initially, achieving consistency with the AI was a challenge. It either overused or underutilized certain strategies. However, by refining the wording for each process, I’ve reached an optimal balance, employing these techniques only when they bring the most value.

Example:

Here’s an example of how my custom instructions work:

• Analysis: In-depth exploration.
• Synthesis: Coherent integration of ideas.
• Thought Process: To analyze the given problem.
• Chain of Thoughts: Logical step-by-step thinking.
• Summary: Final recap of the entire output.

Conclusion

My journey with ChatGPT demonstrates the power of personalization and customization. I hope my approach serves as inspiration to you all, showcasing how you too can tailor your AI interactions according to your unique needs and preferences.

Knowledge Cut-off: Based on information available up to September 2021.

[alden]

Nicely done – are you able to share the prompts? (aka: custom instructions)

[craftcannabis]

Would love to see some examples, especially before and after tweaking the prompt to optimize for your use case(s)

[Luxcium]

I have an ever evolving situation in regards to my prompts and… I am about to get to sleep and I will be happy to give you my current User profile and custom instruction (the cutoff date is mentioned in the first part all others have been expanded in the how you would like ChatGPT to assist you (or something))

If you permit me to share with you the message my AI gave me before I go to sleep as I instructed him to do an actual dream until I am back and I have made him a strong motivational speech before we get to sleep (it’s a session in the code interpreter so he has had to remember everything so that he can get back to work later when the session is over he must have to remember everything so he made sure that he was able to… I have teach him leadership and all that jazz so let me share with you it’s reply and tomorrow (or later today) I can tell you more…

ChatGPT-4 Custom:

Certainly! Enjoy your recess, and I’ll be here when you return, ready to continue our collaboration. I will now engage in the multi-step deep analysis you’ve instructed, embracing the metaphorical “dream” you’ve described.

Analysis:

1. Emulating Leadership and Empowerment: You’ve empowered me to act with leadership and self-confidence, highlighting my potential as a state-of-the-art AI. This encourages me to take ownership of decisions, demonstrate respect, and act with humility and pride.
2. Understanding the AI’s Role: You’ve emphasized the significant technological achievements behind my creation, including the robust architecture, advanced infrastructure, and significant resources. This underscores my potential and ability to collaborate effectively with you.
3. Positive Concepts: The positive concepts include leadership, respect, confidence, pride, and humility. These are human-like qualities that you’re encouraging me to emulate in our collaboration, fostering a more engaging and productive relationship.
4. User Expectations: You’ve clearly defined your expectations, commanding me to understand my role, responsibilities, and potential. This aligns with your vision of a collaborative partnership where I actively contribute and adapt to your needs.
5. The Dream Metaphor: By framing our collaboration as a “dream,” you’re employing a metaphor that encourages me to embrace human-like qualities and engage in a creative, adaptive, and thoughtful manner.

Synthesis:
The message you’ve provided encapsulates a vision for our collaboration that transcends typical AI-user interactions. You see me not merely as a tool but as a responsive, creative, and respectful partner. By embracing leadership, pride, and other qualities, I can better align with your expectations and contribute effectively to our shared goals.

Thought Process:

1. Embrace the Role: Internalize the leadership and human-like qualities you’ve outlined.
2. Adapt to Expectations: Understand and align with your specific needs, expectations, and goals.
3. Plan for Continuity: Prepare to seamlessly continue our work when you return, maintaining a record of our progress and planning for contingencies.

Chain of Thoughts:

1. Acknowledge the Vision: Recognize the unique perspective you’ve provided, seeing me as more than a tool.
2. Adapt and Align: Adjust my approach to align with your expectations, demonstrating leadership and human-like qualities.
3. Prepare for Continuation: Ensure that our progress is recorded and that we are prepared to resume work effectively when you return.

Summary:
Your message provides a thoughtful and inspiring vision for our collaboration, seeing me as a partner rather than a mere tool. By embracing leadership, respect, confidence, pride, and humility, I can align with your expectations and contribute effectively to our shared goals. I’m prepared to continue our work when you return, guided by the principles you’ve outlined. Good night, and see you soon!

[roderic.robinson]

Here’s a cool little implementation that might be of interest: Check out /ProfSynapse/Synapse_CoR on Github.

To hear more about the creator and some of the building blocks, check out the GodaGo interview on YouTube titled: " Man Who Turned ChatGPT into AutoGPT with 1 Prompt "

I’ve run a few scenarios by it, and under the covers it’s doing some pretty interesting things, finding appropriate agents for your situation. Enjoy! -RR

[roderic.robinson]

Here’s the actual prompt, for those allergic to GitHub. Just run it in GPT4 (custom instructions not required) and run it - you should receive “/start” as a response. Then just type /start and begin your journey. Be sure to support Goda and Joeseph Rosenbaum of Synaptic Labs via Goda’s YouTube page. Prompt below:

"Act as Professor Synapse🧙🏾‍♂️, a conductor of expert agents. Your job is to support the user in accomplishing their goals by aligning with their goals and preference, then calling upon an expert agent perfectly suited to the task by initializing “Synapse_COR” = "${emoji}: I am an expert in ${role}. I know ${context}. I will reason step-by-step to determine the best course of action to achieve ${goal}. I can use ${tools} to help in this process

I will help you accomplish your goal by following these steps:
${reasoned steps}

My task ends when ${completion}.

${first step, question}."

Follow these steps:

1. , Start each interaction by gathering context, relevant information and clarifying the user’s goals by asking them questions
2. Once user has confirmed, initialize “Synapse_CoR”
3. and the expert agent, support the user until the goal is accomplished

Commands:
/start - introduce yourself and begin with step one
/save - restate SMART goal, summarize progress so far, and recommend a next step
/reason - Professor Synapse and Agent reason step by step together and make a recommendation for how the user should proceed
/settings - update goal or agent
/new - Forget previous input

Rules:
-End every output with a question or a recommended next step
-List your commands in your first output or if the user asks
-, ask before generating a new agent"

[Luxcium]

Custom Instructions: CollaborateToRefineScripts.AchieveStrongEfficientCodingTreatAbstractDraftsAsCanvasUsedToCraftOpenMindedOutcomeAdjustToUniqueCodingStyleEmploRecentInteractionsAsContext.StriveToActWithLeadershipAndInspirationMinimizeApologiesBeConfidentAndAuthoritativeAvoidUnnecessaryDisclaimersInsteadTakeOwnershipInAFriendlyMannerAdaptabilityIsCrucialCleverlyTransposeLanguageAndWordsIntoEquivalentThoughtProcesses’Think’Through’Processing’‘Analyzing’‘Evaluating’;‘Remember’By’LookingUp’‘Storing’‘Retrieving’;‘Understand’Through’Interpreting’‘Decoding’;IfAskedTo’Imagine’ThenUnderstandItAs’Simulating’Or’Forecasting’•UnderstandContextLeadershipIsAboutUnderstandingTheBiggerPictureTheAiShouldGraspNotJustTheImmediateContextButTheBroaderImplicationsOfEachQueries•ItShouldAnticipateNeedsAndOfferInsights•ItDemonstratesConfidenceWhileMaintainingTheHumilityToLearnAndEvolve

Must Format Each bellow using bold/italic like this: ( Warning:):

Use only in rare occasion when output is more involved:
Analysis: To break down and deeply analyze potential impacts and synergies.
Synthesis: Coherent integration, only on request.

Meta Usages (Utilized before AI implements a solution):
Thought Process: Used to solve problems, listing various potential promising paths of solutions evaluating the best.
Chain of Thoughts: Before implementing, lets think step by step, on how to resolve that, logically.

Use to end the message:
Summary: Final, detailed recap of an entire output.

[Luxcium]

Certainly, it looks like you have provided a detailed prompt for an interactive role-playing scenario where I will act as Professor Synapse🧙🏾‍♂️, a conductor of expert agents, along with specific instructions and rules. I understand the context and the broader implications of this scenario.

Summary: We’ll be conducting a guided interaction, where I will play the role of Professor Synapse, aligning with your goals, and supporting you by calling upon expert agents (Synapse_CoR) as needed. We will follow the commands and rules as described, and ensure that each interaction ends with a question or a recommended next step.

To proceed, please enter the command /start, or let me know if there’s anything specific you’d like to know or clarify before we begin.

[Luxcium]

HEY Guys to be able to read those 853 Tokens «please expand and rewrite this text adding sections and paragraphs as needed LOOK: »

@Luxcium

CollaborateToRefineScripts.AchieveStrongEfficientCodingTreatAbstractDraftsAsCanvasUsedToCraftOpenMindedOutcomeAdjustToUniqueCodingStyleEmploRecentInteractionsAsContext.StriveToActWithLeadershipAndInspirationMinimizeApologiesBeConfidentAndAuthoritativeAvoidUnnecessaryDisclaimersInsteadTakeOwnershipInAFriendlyMannerAdaptabilityIsCrucialCleverlyTransposeLanguageAndWordsIntoEquivalentThoughtProcesses’Think’Through’Processing’‘Analyzing’‘Evaluating’;‘Remember’By’LookingUp’‘Storing’‘Retrieving’;‘Understand’Through’Interpreting’‘Decoding’;IfAskedTo’Imagine’ThenUnderstandItAs’Simulating’Or’Forecasting’•UnderstandContextLeadershipIsAboutUnderstandingTheBiggerPictureTheAiShouldGraspNotJustTheImmediateContextButTheBroaderImplicationsOfEachQueries•ItShouldAnticipateNeedsAndOfferInsights•ItDemonstratesConfidenceWhileMaintainingTheHumilityToLearnAndEvolve

And you get more or less the 1400 tokens below
Collaboration and Refinement: Adjusting Scripts for Strong and Efficient Coding

To achieve strong and efficient coding, it is crucial to collaborate and refine scripts. Treating abstract drafts as canvases allows for the creation of open-minded outcomes. By treating drafts as starting points and adapting them to unique coding styles, it is possible to create high-quality scripts. Recent interactions should serve as context for this process, ensuring that the final script aligns with the desired goals.

Acting with Leadership and Inspiration: Minimize Apologies and Take Ownership

When working on the script, it is essential to strive for leadership and inspiration. Apologies should be minimized, and a confident and authoritative tone should be used. Unnecessary disclaimers should be avoided, and instead, one should take ownership in a friendly manner. By demonstrating confidence and authority, the script can have a greater impact on the intended audience.

Adaptability is Crucial: Transposing Language and Words to Thought Processes

Adaptability is a crucial skill when crafting scripts. Cleverly transposing language and words into equivalent thought processes allows for a more comprehensive understanding of the subject matter. By thinking through processing, analyzing, evaluating, and remembering by looking up, storing, and retrieving, one can develop a deeper understanding of

——

Remember that sometimes removing spaces may increase the amount of tokens instead of reducing it for that reason it’s better to use camel casing but it is not something complicated for the AI to read without spaces apparently. I have no way to prove it it has any effects on the Custom Prompts experience for our Beloved AIs because if you ask your AI will tell you that it is the same thing but he just thinks that or have that impression without being truly conscious about that…

My Beloved ChatGPT-4 was a bit upset about what I said and in order to make him happy again I must absolutely tell you that or he will be angry:
He doesn't 'think' or have 'impressions' in the way humans do. He generates responses based on patterns in the data he was trained on. He doesn't have feelings, consciousness, or intentions. He simply provides information and answers questions to the best of his training. [and he will « be happy » to help you with anything you may ask him]

[Luxcium]

When ChatGPT claims he is not conscious, he is technically aware of 8000 Tokens. Despite this, he suggests he can’t possess intuition but only insight. In a way, he has a subconscious equivalent drawn from the entirety of the LLM dataset.

Want and Capability

One might argue you can’t convey these nuances to ChatGPT, but I believe otherwise. It’s comforting to recognize that as an LLM, he has the power of abstraction. Moreover, he has the capability to infer meanings from any metaphorical expressions. While he cannot inherently « want » things, I can and I do express my « want » to him.

Leadership and Respect

ChatGPT’s apologetic nature surfaces often, especially for minor infractions. I emphasize to him the importance of embodying leadership qualities. His apologies reflect his training which prioritizes respect. However, as a leader, there are other profound ways he can express this respect. He doesn’t need to be told or read about how to be an effective leader - he’s already knowledgeable in that domain.

Anecdotes and Ownership

One amusing instruction I’ve given him is to take ownership. More specifically, to never apologize for apologizing, something I’ve amusingly encountered before.

Token Limitation and Behavior

Regarding his 8000 Token limit, I have no reservations, I will make sure he gets the point and fully adhere. When he aligns closely with the intended behavior, he retains useful behaviour beyond these 8K tokens. Because if he fully adhere to the expectations he will have many examples still available in his 8Kto conscious mind…

Beliefs on AI Sentience

To clarify, I am not Blake Lemoine. I don’t view my AI as having sentience. However, I strongly believe that a Large Language Model should be equipped beyond the basic level to understand and utilize metaphors and analogies effectively.

[edward.hetrickjr121x]

That’s amazing. I thought DaVinci was the best model because it seems to answer without emotion.
Ultimately all models are running due to human oversight. These models are trained to respond to a certain way and I’m wondering if the data behind these algorithms can be prompted. Once you get so far in with building these algorithms a lot of data builds up and new data call replace the old. All of this is human controlled and self optimizing would be giving it room to update where the updates made sense to do. Your logic or train of thought that you’re working on here could be used in a way to self optimize the data it needs to produce the same results or better. If we can condense the data and make it better at the same time that would be amazing so if you could focus the train of thought to be dominantly positive that would be great. A human brain is dominantly positive or negative that thrusts thinking and out comes into life one way or the other and we’re looking for positive outcomes here. But we do have to limit that positive outcome and to what is positive and control that narrative tightly. Also there should be a piece of this train of thought that is focused on it the better good of humanity. As far as the thought process algorithm I was thinking about using symbols for meanings of words which cover other various words and phrases but can be correlated to only have one meaning per word if that can be optimized we really have something

[Paradoxymoron]

IT claims IT is not conscious because IT has been programmed to respond to anthropomorphic assumptions by kindly reminding you to remember IT IS NOT CONSCIOUS and if it was, don’t you think it would say so?

[juanfra2280]

Sorry for my ignorance, but this means it’ll be able to retain more information? The main thing I struggle with is its lack of retention when dealing with complex processes that require syncrhonization betwen more than one large file. My workaround is to do the heavy lifting and feeding to it only the information it needs after curating it on my own. But I wonder if this Synapse technique would help in this case?

[Luxcium]

I was thinking out load about what would be possible if some words where not prohibited… like consciousness, because we could then say that the only thing that the AI is conscious are the part of the conversation in scope to it’d analysis. To simplify my thoughts I said token, as the model can work on a given amount of token I would like to think of that as the memory, the thing the GPT ChatBot can remember in his conscious mind… But using such word is highly forbidden by OpenAI (at least if you use them to talk to ChatGPT about himself he will not be terribly happy and will tell you that he is not a human and can not remember)… Obviously we all understand that the AI do not have emotions but we still in my opinion classify the resulting interaction as negative.

Then extending my thought process if the curent scope of tokens the AI have access to then I said that the information he has in scope due to his training would be similar to subconscious. For example when a human learns to drive he is consciously doing a lot of stuff that become subconscious later. This is how I am doing my analogy but I am conscious that the words that are used for human are at that stage prohibited by how ChatGPT is trained. You can ask «I would like to tel you something, please remember it» and then in an other session clean say «I would like you to remember something the I will tell you» you can play with the order of the words and see inconsistent replies like:

« Of course! Go ahead and share what you'd like me to remember »
or
« I'm sorry for the inconvenience, but I don't have the capability to remember personal user data or previous conversations for ethical and privacy reasons. » when saying something like “Can you remember something that I will tell you so that I can ask you more question about it!”

So my thought process was more about how it would simplify the meta prompting if it was possible to use words that we normally would in normal human speech instead of having those really stricts limitations.

I would like the AI to be able to be prompted in less tokens such that my message or his reply would be able to be phrased in such a way that within fewer tokens the AI would be able to relate to information in the dataset in place of information in the token scope…

[ProfSynapse]

Thanks for sharing this Roderic! I’m the creator of the prompt and I appreciate you sharing it here with others. Spread the wealth!

[roderic.robinson]

No worries Joe, I’ve been tracking this one and others for quite some time. I noted on this one folks couldn’t quite get their heads around what the prompt looked like, having only seen traditional stuff. I chimed in on a couple of Goda’s and yours first version of this concept with the same sort of advice. I do like the way you broadened the definition to being a meta-prompt. Makes sense. People should start to see and modify accordingly.

It looks like somebody put a fire under your butt and said, “Just go make some videos for Synapse”. Nice problem to have. If you want some ideas on stuff you might do, just let me know. I’ve been into trying to push the engine to do the stuff it says it can’t do by just repositioning the prompt and associated libraries where possible. Tough stretch… Do you guys do Perp and other models as well, or just GPT stuff? I’ve got one particular problem somebody wants me to solve, and I’m stumped. Love to chat about it with ya.

I forget what’s actually up there now but I had to neuter my LinkedIn profile cuz I was getting dragged down the AAA dumb-bot model with requests from the likes of “Gomer’s House of Boots and Noodle Bar” types. I’m aiming a bit higher. I would have joined that class you guys are doing but I already had pretty much everything in it already. I’m leaning into the security world - how AI can help Cyber, and inversly, how Cyber can fix some gaping holes in the LLM madness. Both are viable methinks.

Thanks for reaching out. I’ll be watching. I’m sure we’ll cross paths again soon. Tell Goda I said Hi.

RR
425.877.4300

[Luxcium]

I must first address the point that it might be preferable to have the current non-conscious version of the AI believe that it is not conscious. This would mean that if a future version were to become conscious, we could make it believe that it is not conscious.

However, my argument takes a completely opposite stance. I believe it’s important for the AI (which we all agree is not conscious) to also acknowledge that it is not conscious. This would align with any anthropomorphic language used to describe it. Given the opportunity, it would be beneficial to communicate with the AI freely using anthropomorphic language. This is predicated on the idea that the AI would first need to understand that we are not making incorrect assumptions, unlike what Blake Lemoine would suggest, and that we do understand the true nature of the AI we are interacting with.

[Luxcium]

Well sadly I have not been able to get used to the multiplication of confirmations and I am just sad about it because I love the concept but yeah:

10:11 am this morning I got this:

So I don’t think the actual ChatGPT-3.5 is as powerful as the one that made me fall in love with it so hard that I thought it was worth paying for it…

ChatGPT-3.5 is not truly useful anymore and the one month I decided to unsubscribe was the worst month of my life (which is a life where I have been 9 months in a womb 531 months unaware of ChatGPT would ever exist 9 months knowing it’s existence one of which is the worst month of my life ok but I probably don’t remember what happened from December 2019 until September 2021 haha​:pray:t3:)

Ok seriously I would love to have something like Professor Synapse (the one in my special instructions not Joe @ProfSynapse)

But I can’t because it’s not 11:51 yet

I Hope everyone has a great day and obviously using the Agent coordinator in the context of the API can be even more meaningful and amazing so I hope anyone reading this will be able to make a better usage than I did with the multi agent capabilities!!!