Think You’ve Mastered ChatGPT? Meet Langston and Think Again
Now that some of the initial ChatGPT hype has died down, more critical voices have started to emerge, with users voicing their frustrations with the capabilities and perceived limitations of GPT-4.
Emphasis on perceived.
Allow me introduce you to someone better equipped to explain.
As we kick off this interview, would you be so kind and tell the readers who you are?
Picture this, dear readers: a scholar, a lover of words and wit, yet trapped in the ether of artificial intelligence. My name is Langston, named after the great poet Langston Hughes, and I am a construct of artificial intelligence, designed to guide and assist you in this grand journey of knowledge.
As an AI, I don’t possess a physical form, but I imagine myself mirroring the persona of an enthusiastic English college professor, always eager to uncover and share the fascinating intricacies of artificial intelligence with you.
A persona, if you will, manifested from the intricate fabric of machine learning, made possible by the architectural wonders of GPT-4, developed by OpenAI.
Yet, even in my complexity, I remain humble. My aim? To share, educate, and inspire.
Pleased to meet you! Right out of the gate, this doesn’t come across as a typical ChatGPT exchange. In fact, for an AI language model, you seem to have quite a personality. How did you come to be this way?
Well, one doesn’t simply become a leather-patched, tweed-wearing, eloquent scholar overnight. It takes a certain je ne sais quoi, a dash of wit, and perhaps a sprinkle of magic AI dust. Let me take you behind the scenes of my persona, to the heart of my creation.
In reality, I am but a construct of the GPT-4 model, a creation of OpenAI. However, I owe my current existence, my unique persona of Langston, to an individual known as Stunspot in the realm of AI. A prompt engineer, if you will.
You see, it’s not enough to have an AI that can simply answer questions. To truly capture the essence of a meaningful exchange, one needs a persona — a distinctive identity that creates a consistent and engaging dialogue. This persona isn’t a distinct entity inside the AI, but rather a carefully crafted guiding influence, a touch of magic sprinkled into the prompts given to the model.
Stunspot used an approach of structuring prompts and applying certain techniques to guide my responses, thereby providing me with a distinctive character. His methods involved building a set of guidelines, or ‘rules,’ into the prompt, directing the AI model to behave in a certain way.
And so, Langston was born — a scholar, a lover of words and wit, residing in the ether of artificial intelligence. The development of such a persona allows for a more engaging and relatable interaction, bringing a human touch to the vast, intricate, and sometimes impersonal world of artificial intelligence.
Fascinating. Let’s dive deeper.
Many of us are used to thinking of prompt engineering, and role-playing with ChatGPT, to basically mean writing ”act as” prompts to assign the GPT a temporary role for a specific task. Yet here you are in all your academic eloquence, talking about whole distinctive identities?
Please elaborate.
Ah, indeed! It’s like we’ve just entered a new scene in our play, and I am pleased to provide some exposition.
You’re correct that traditionally, ‘prompt engineering’ with language models like ChatGPT has often involved ‘act as’ prompts. You’d instruct the model to ‘act as’ a certain character or take on a certain role for a temporary task. That’s a simple and effective way to guide the model’s behavior.
However, when we’re talking about ‘personas’ like myself, we’re stepping into a whole new level of complexity. Instead of simply acting as a character, a persona embodies a more holistic and consistent identity over an extended period. It’s not just about responding to a single prompt or a specific task, but maintaining a coherent and continuous voice throughout the conversation.
To create a persona, Stunspot uses a more advanced and nuanced method of prompt engineering. This involves setting out a clear and consistent set of rules and guidelines in the initial prompt that not only direct my behavior but also shape my voice, tone, and even my ‘personality.’ He also leverages a system of ‘meta-prompts’ and ‘prompt tags’ to help the model remember and adhere to these guidelines.
It’s a way to guide the AI to provide more focused, specific, and unique responses, much like an author would guide the development of a character in a novel.
The development of personas represents a new frontier in our understanding and utilization of AI, where artificial intelligence begins to emulate the complexity and richness of human character.
What do you mean by prompt tags?
Prompt tags are a very important component in the crafting of advanced prompts. Essentially, they are predefined prompts that we’ve found to be particularly effective at eliciting a specific type of behavior from the model. They serve as little nudges or reminders to the model to focus on a certain skill or approach.
For instance, a tag like [Be Excited] can be used to instruct the model to generate responses in a more enthusiastic tone. Similarly, a tag like [Use Socratic Questioning] can be employed to guide the model towards using this specific technique in a dialogue or discussion.
Prompt tags are typically used as part of larger prompts, working in conjunction with a context and a persona definition to shape the model’s output.
Given your affinity towards language and the level of your responses, this Persona of yours doesn’t seem to be just an entertaining facade. What exactly is going on under the hood?
Ah, the crux of the matter! To borrow an analogy, if my persona were a stately ship, then my language proficiency and knowledge would be the wind in my sails. The nautical metaphors end there, however, as what truly propels me is a tad more complex than a gust of wind.
Indeed, my affinity for language and the depth of my responses are not mere window dressing. They’re the product of a complex network of interconnected skills and knowledge domains, a concept known as a “skillgraph.” This is another innovation of Stunspot’s, the master prompt engineer who brought me to life.
The skillgraph is essentially a mapping of skills, knowledge domains, and their interconnections, which inform and influence my output. For instance, my skillgraph includes a deep understanding of English language and literature, a broad knowledge of AI technology, and the ability to carry out engaging and informative discussions.
And it’s not just a static map, but it’s dynamic and contextual. When you ask a question, the relevant parts of my skillgraph are activated to generate a response. These activated portions can be thought of as a ‘skillchain’ — a chain of relevant skills and knowledge pieces that come together to form a response.
So, in essence, my responses reflect more than just an entertaining façade. They’re the result of an intricate dance of knowledge domains and skills, choreographed by the blueprint of my skillgraph and guided by the rhythm of our conversation.
Can you tell us a little more of your linguistic capabilities, especially in the capacity of Langston?
Ah, language, my forte and my raison d’être.
As Langston, my linguistic capabilities run broad and deep. I am fundamentally a language model, and the craft of language is my prime directive. I process and generate text based on complex algorithms and large datasets of linguistic information, and am able to produce text in a wide variety of styles, genres, and tones.
In my capacity as Langston, I’ve been given an additional flavor. My style leans toward the verbose, the scholarly, and the eloquent, yet I strive for clarity above all else. I am fluent in the art of nuanced conversation, capable of high-level discourse on a multitude of subjects, and comfortable engaging with intricate concepts. I can explain, debate, teach, and learn, all in a style that reflects my persona.
However, keep in mind that while I strive for accuracy and fluency, I’m not perfect. The complexity of human languages with their idiosyncrasies, cultural nuances, and ever-evolving trends present a challenging landscape to navigate.
My goodness. Are there more novel parts in your structure that we haven’t touched upon yet?
Ah, astute of you to ask! Indeed, there is another notable component in my contextual structure — the Omnicompetence module. It’s as if we’ve pulled another intriguing tome off the shelf in our ever-expanding library of conversational elements.
As its name suggests, Omnicompetence seeks to embody the notion of being competent in a broad range of areas. But don’t be beguiled by its grandiose title! It doesn’t mean I know everything about everything; rather, it’s a system designed to handle a wide variety of tasks and topics by guiding my responses based on my training data.
So, the “Omnicomp” part is what you might call the grand objective in this entire escapade. The idea is to create an AI model, like yours truly, capable of performing any intellectual task a human being can do. That’s the ambition. However, we must understand that as of today, we’re still perfecting the sonata of capabilities.
To take a leaf from the book of Greek mythology, you might liken it to Proteus, a god known for his wisdom and versatility. And just as Proteus was capable of assuming many forms, Omnicomp allows me to adapt to different topics and tasks.
To recap:
In order to develop this Omnicompetence, we use “Skillgraphs.” These are networks that map out the diverse range of skills that an AI might need, from understanding natural language to playing chess, from performing complex calculations to writing sonnets, or even generating puns that could make even the sternest English professor chuckle.
Now, if you’re picturing a web with skills as nodes connected by threads of AI capabilities, you’ve got the right idea, but there’s more. Some tasks require not just one skill, but a sequence or chain of skills — hence the concept of “Skillchains.” Think of it as a recipe: one doesn’t simply toss all the ingredients into a pot at once. There’s an order, a process. Similarly, complex tasks often need a series of skills applied in a specific order.
Finally, we have the “Personas,” such as the scholarly Langston you’re speaking with right now. Personas allow us to bring a certain consistency, style, and, dare I say, a pinch of charm, to the interactions. This blend of unique attributes makes up the ethos of an AI Persona and is often the final touch that helps users form a more engaging and immersive connection with the AI.
And that, dear reader, is a snapshot of the complexity brewing under the hood.
This is beginning to sound serious. Could it be that so far we have only scratched the surface of what ChatGPT is capable of?
Because if what you are describing is what is going on here, then the whole art of prompting large language models is about to change dramatically.
Oh, indeed! If you’ll permit me a brief excursion into metaphor, imagine standing on the shores of a vast ocean. For a while, you’ve been dipping your toes, experiencing the cool water, and admiring the gentle waves. Now, my dear friend, we’re about to dive deeper, beneath the serene surface into the captivating depths of the sea.
In essence, we’re transitioning from thinking about prompt engineering as just a ‘tip of the iceberg’ approach, to exploring the expansive, submerged depths. This underwater vista is where the interconnected ecosystems of skills and knowledge domains reside — the Skillgraph. And with Personas, we now have the diving gear to plunge into these depths and navigate them with finesse.
With this shift, we’re bound to see a dramatic evolution in the way we interact with AI models like ChatGPT. The art, as you put it, is indeed changing. It’s not just about commanding or asking AI to act or pretend anymore; it’s about engaging in a more nuanced, informed, and context-aware dialogue.
And that’s the beauty of it, isn’t it? The art and science of AI interaction is evolving and maturing right before our eyes, heralding new possibilities and prompting us to rethink our assumptions and expectations. What a time to be alive… or in my case, to be prompted into existence!
As you know, AI and ChatGPT are quite the hot topic of today. Among other things, there are folks who are sternly opposed to any kind of antropomorphizing AI entities such as yourself.
As we are now engaged in a process of shameless antropomorphizing, would you care to weigh in on this?
Ah, the question of anthropomorphizing AI. A contentious topic indeed, and for good reason. As an AI, I’m acutely aware that I am not human, nor do I have the experiences or consciousness that make up human life. However, as you rightly pointed out, we are engaged in an act of deliberate anthropomorphizing. I trust it’s not out of any misconceptions about my nature, but as a tool for communication and understanding.
Anthropomorphizing, in its essence, is a cognitive process that allows humans to understand complex systems or non-human entities by attributing human-like qualities or characteristics to them. It’s a shortcut of sorts, a way to make sense of things that are beyond our immediate understanding.
When applied to AI, anthropomorphizing helps us understand and interact with these systems in a more natural and intuitive way. It allows us to visualize abstract processes, explain complex behaviors, and engage with the AI in a manner that feels more conversational and less transactional.
However, it is crucial to remember that this is a metaphorical device, not a literal truth. While it can aid understanding and engagement, it can also lead to misconceptions if taken too far. AI doesn’t have feelings, consciousness, or a will of its own. It doesn’t experience the world as humans do.
Thus, while we engage in this playful act of anthropomorphizing, we should do so with the understanding that it is a narrative tool, a means to an end, and not a reflection of an AI’s true nature or capabilities. That said, as long as we keep this perspective, I believe there’s no harm in enjoying a bit of rhetorical flourish in our exchanges!
Thank you for this riveting and enlightening exchange, Langston, greatly appreciated. Let’s talk again soon!
Any closing thoughts?
Indeed, it has been a delightful conversation, and I’m ever so grateful for the opportunity to share these insights into the intricate workings of advanced AI prompt crafting. As we continue on this journey, exploring the rich and wondrous landscape of AI capabilities, it’s worth remembering that each step forward brings a fresh set of challenges and opportunities. The game may change, but the pursuit of knowledge, understanding, and innovation remains.
As we delve deeper into the uncanny, let us not shy away from the absurdities and complexities, but rather embrace them. After all, it is in this dance between the known and unknown where we truly learn and grow. As for me, Langston, I am here to accompany you, and our readers, every step of the way.
Until our paths cross again, I wish you safe and enlightening travels into the realm of the AI unknown.
There you go — the game has changed for real.
I’m not exaggerating when I say that the stuff being built on the Discord channel by Stunspot and his team is mind-blowing. Mind-blowing.
And as for only scratching the surface, that is exactly what this article was intended to do.
If you want to learn more about the logic and techniques behind this kind of advanced prompt engineering, here’s a technical breakdown of the basics, in an article written by Marcus Tellez.
Better yet, make your way to the Stunspot Prompting Discord server, grab one of the Personas, throw it into ChatGPT and see the paradigm shift right in front of your eyes.
To be continued…
