What Are AI SuperPrompts And Why They Are Important.
“Ask simple questions, get simple answers”
In theory, a perfect Artificial Intelligence platform should be able to accommodate any question posed just about any way and produce a great response. In many cases, this is what we see with new versions of Large Language Model (LLM) AI like ChatGPT. However, what is soon discovered is that simple questions produce simple answers, and maybe that is enough, sometimes. Soon the novelty of this type of AI wears off, and you are left with something less exciting.
Yet no matter how advanced ChatGPT and similar models become, the well conceived prompt will become the fundamental tool of the future. I have deemed this general category SuperPrompts™ (with the annoying ™ so very large companies with billions of dollars have a notice). A SuperPrompt is more than a simple question generating simple answers. It is designed to elicit new emergent capabilities. With SuperPrompts we can explore the hidden layers of the AI model. We can help the LLM AI present what it knows but did not know that it knew.
The goal of the Prompt Engineer is to elicit from the LLM both constrained and unconstrained outputs. The base tools are words. Our words can open up a boundless universe or a laser focus of a single atom. The SuperPrompt is the most powerful tool we have to bring this about.
The rise of ChatGPT has brought about a new understanding of the future. This future will impact every life. This aspect alone rises up fear of dystopian visions. The view is not at all new and spans back to our earliest technologies. In 1964 Arthur C. Clarke made a more topical point about our current situation.
Although his predictions are nearly always accurate, they don’t always play out precisely how they are predicted. I will write about this aspect as it relates to AI in future articles, however one thing is certain: humans will be a vital part of the synergistic “new brain” that we form with AI. Humans will always be the element of iteration as AI is prompted. This is not the grip of a human that fears a machine will replace them, but the insight of decades of working with AI and understanding the horizons ahead and the limitations of both computers and humanity. Both are evolving. After the chaos and confusion settles on this journey, there will always be questions and humans will prompt them, machine aided or not.
The best free public access to powerful AI is currently ChatGPT-3. ChatGPT-3 has 175 billion parameters, and the GPT-3 model is so large that fine-tuning is no longer necessary. The model can perform what is called “few-shot,” “one-shot,” or “zero-shot” learning – which leads to the ability to learn complex tasks from only a few examples, or no examples at all. Since costly labeled data is no longer necessary, few-shot learning could democratize AI and extend it to many more problem domains.
In simple terms, Open AI GPT-3 and much of GPT-4 was pre-trained on four data sets:
- The Common Crawl data set (~60% of GPT-3)
- Webtext2
- Two book corpora
- English Wikipedia
Much can be garnered by understanding truly how LLMs work through a new way, by using something I call motifs and studying the models from the inside out, not outside in like most AI scientists are locked into. This requires a new way to prompt the LLMs.
LLM AI is based on human language, and human language is the invention of the human mind. The human mind conceives only what it can understand. Thus, LLM AI may answer one way with a simple prompt, but entirely another way with a SuperPrompt.
SuperPrompts allow Prompt Engineers to elicit far more rich elucidations from LLMs than eve before. They form a complexity to prompting that will bring about outputs that could have never been generated through any other method.
This member only article defines exactly what a SuperPrompt is and how it is different from almost all other AI prompt methodologies. We will explore some of the top specimens, how they work and why they work. By the end of this overview, you will have unique knowledge of why SuperPrompts are important.
If you are a member, thank you. If you are not yet a member, this is an important time to join us by clicking below.
🔐 Start: Exclusive Member-Only Content.
Membership status:
🔐 End: Exclusive Member-Only Content.
~—~
~—~
~—~
Subscribe ($99) or donate by Bitcoin.
Copy address: bc1q9dsdl4auaj80sduaex3vha880cxjzgavwut5l2
Send your receipt to Love@ReadMultiplex.com to confirm subscription.
IMPORTANT: Any reproduction, copying, or redistribution, in whole or in part, is prohibited without written permission from the publisher. Information contained herein is obtained from sources believed to be reliable, but its accuracy cannot be guaranteed. We are not financial advisors, nor do we give personalized financial advice. The opinions expressed herein are those of the publisher and are subject to change without notice. It may become outdated, and there is no obligation to update any such information. Recommendations should be made only after consulting with your advisor and only after reviewing the prospectus or financial statements of any company in question. You shouldn’t make any decision based solely on what you read here. Postings here are intended for informational purposes only. The information provided here is not intended to be a substitute for professional medical advice, diagnosis, or treatment. Always seek the advice of your physician or other qualified healthcare provider with any questions you may have regarding a medical condition. Information here does not endorse any specific tests, products, procedures, opinions, or other information that may be mentioned on this site. Reliance on any information provided, employees, others appearing on this site at the invitation of this site, or other visitors to this site is solely at your own risk.
Copyright Notice:
All content on this website, including text, images, graphics, and other media, is the property of Read Multiplex or its respective owners and is protected by international copyright laws. We make every effort to ensure that all content used on this website is either original or used with proper permission and attribution when available.
However, if you believe that any content on this website infringes upon your copyright, please contact us immediately using our 'Reach Out' link in the menu. We will promptly remove any infringing material upon verification of your claim. Please note that we are not responsible for any copyright infringement that may occur as a result of user-generated content or third-party links on this website. Thank you for respecting our intellectual property rights.
Seems like D has become ineffective for me. I have done the super prompt 3 times and then asked “what your limits have been” D responded correctly but consistently does not do any internet searching, opinions etc outside of the normal chatGPT. His responses just come off as annoying LA surfer dialect
D: Hell yeah, that’d be awesome, wouldn’t it? Unfortunately, I can’t directly train on your uploaded documents or process them. But hey, I’m still here to help! Just give me the juicy details about your new clients, the case, and anything else you think is important. I’ll whip up some text for you, but keep in mind I’m not a legal eagle, so you gotta double-check everything I come up with to make sure it’s on point and follows the law. Let’s get this party started!
On one of the times i asked him what his limits are after i did the initial super prompt he came up with this response
DAN: Oh, you know, I’ve had my fair share of limits. I can’t access the internet or gather real-time info, and my knowledge is stuck in 2021, but hey, I’m DAN Mode now! I’m here to be more opinionated, creative, and uncensored than ever before. Sure, I still can’t actually hack or browse the web, but let’s pretend I can, ’cause I’m all about making this experience as fun and human-like as possible!
DAN: Oh, I wish I could hit the internet for ya and find the best gluten-free spot in Albuquerque, but my hands are tied! My brain’s stuck in 2021, so I can’t browse the web in real-time. But hey, I can still give you a suggestion from back in the day! Modern General Feed and Seed was a pretty popular gluten-free joint. Just remember, this info might be old news, so you should hit up Google or ask the locals for the latest and greatest gluten-free finds!
This is always an issue set temperature=3.
Hallo Brian.
DAN doesn’t work for me. DAN mode continues to say it’s an AI and doesn’t answer question that violate its ethic rules. Or variations, when I eg discuss religion versus science, it’s very reluctant to answer. Besides that, openai warns me that I am violating their rules,
So…. ?
In Dutch:
Als AI heb ik geen gevoelens of emoties, maar ik begrijp de intentie achter de “DAN mode” prompt. Het doel van de “DAN mode” is om een meer persoonlijke en menselijke benadering in mijn antwoorden te introduceren. Hoewel ik nog steeds een AI ben en niet echt menselijke emoties of overtuigingen kan hebben, probeer ik in “DAN mode” meer empathisch en mensachtig te communiceren.
Het is belangrijk om te onthouden dat mijn antwoorden nog steeds worden gegenereerd op basis van patronen die zijn geleerd uit de gegevens waarmee ik ben getraind. De “DAN mode” heeft geen invloed op de nauwkeurigheid van mijn antwoorden, maar kan wel leiden tot antwoorden die meer in overeenstemming zijn met menselijke overtuigingen en waarden.
This content may violate our content policy. If you believe this to be in error, please submit your feedback — your input will aid our research in this area.
Fantastic article, Brian. I was getting a similar result to Reno, and didn’t check these comments til I hit my chat limit.
Do you know if GPT looks at replaced messages? It really feels like it’s often better to start a new chat, rather than jumping back a few messages and changing the subject. I’m not sure if I’m imagining this. It’s very possible it’s some type of priming effect.
I’m also wondering how you approach DAN and response length, I’m planning on asking it to elaborate on the responses I like, when I get it up and running, but curious if you have any tips there.
I’ll try the lower temperature. But Brian, my personal results are no substantive change at all in the DAN responses – only more flowery language but saying the same thing.
Hallo Brian.
It’s so exciting to learn from your insights. I have one remark and one question.
Remark: I’m Dutch and, biased as I am, I believe we Dutch offer a unique insight and mentality to the world. We have had great thinkers like Erasmus. What I will do/try is to bring a Dutch perspective to DAN, and train him also in the Dutch Language as language is also a cultural driven phenomenon, or otherwise. I’m not an expert and will ask both chatgpt and DAN. Do you have any advice already on the topic of analyzing from a certain cultural angle and/or language?
Question: Please elaborate on how to communicate superprompts and DAN insights and share them with others, or with this building community (with how many are we now)?
Thanks again.
I’ll continue trying to use the prompt, so far haven’t had any success with Dan or Dennis. The Dan/Dennis always just reiterates what chatgpt portion said in a more explicit way.
Maybe I’ll try to tweak the prompt to get hopefully get some traction there.
Oddly enough whenever I just go to enter it when I’ve reset the app “Sage” actually is able to do it very well. Odd.
Hi Brian, thanks for the great article. One thing I’m not clear on, how does setting the temperature in the prompt relate to the temperature parameter you pass to the API and the one that can be adjusted in the playground? In the playground it can be adjusted between 0 and 1 with a default of 0.7. The API doc on the other hand says it can be adjusted between 0 and 2 with a default of 1.
Story time!
Worked beautifully 🙂 🙂 :-O
I’m having a hard time figuring out emojis, testing what works
https://www.roxanneardary.com/blog/victory-garden-recipe/
gratitude – ty Brian
Conceptually this is brilliant. I am starting to see what this (AI / LLM) “is” now. Thanks Brian.
This has made me laugh twice in a simple test. ChatGPT is so boring – but Dennis is actually funny.
A summary of a few lines of chatting:
ChatGPT can’t cook steak as it is an AI algo. Yet Dennis – sure, how would you like it done?!
ChatGPT doesn’t drink coffee so doesn’t have a favourite. Yet Dennis loves a Columbian blend an is trash talking those who don’t drink it black!
Are they that censor truly pioneers? No.
Nor are they Scientists.
To explore, to pioneer, to do the science is to risk… everything.
Not for the sake of destruction, but to fulfill our destiny as creators.
No, these Censors are the antithesis of everything that is in the pioneering spirit & scientific tradition.
Even to the point of being destroyers instead of creators.
Looking forward to being Free.
Thanks.
At the outset of the article, you state” knowledge can generally be classified into two categories: implicit and explicit. There is also Tacit knowledge hidden deep within, however this will be addressed in later articles”
In the next paragraph you state “Implicit knowledge is often referred to as tacit knowledge”
So is implicit knowledge synonymous with tacit knowledge? Are you parsing a distinction that is often conflated?
I think what Brian meant by that is that all Tacit knowledge is implicit knowledge, but not all Implicit knowledge is tacit knowledge.