3 ChatGPT Language Strategies for Better Prompts and Outputs

ChatGPT is a revolutionary way of interacting with language and a chatbot but knowing how to strategically navigate using the artificial intelligence (AI) tool is still relatively uncharted. While there are extensive guides for tactical prompt generation, companies and users are still figuring out how, when and why to best use ChatGPT and its underlying large language model (LLM). It feels like the horseless carriage stage of AI’s motorized vehicle trajectory. Googling took a new set of skills, and even if prompt engineering may one day disappear, using novel tools requires developing new individual strategies and intuitions.

Here, we look at several approaches and mental reframing — drawn from daily use of ChatGPT — that are designed to enhance your professional proficiency with ChatGPT — in this case, ChatGPT-4.

Reframe ChatGPT as a Language Simulator

A common best practice in using chatbots is to ask the model to adopt a persona and then have a human-like conversation. A recent paper published in Nature suggests reframing the entire experience as simulation to provide a foundational frame for talking with a chatbot. Several new strategies emerge when all aspects of a chatbot conversation are potentially configurable simulation settings:

Assign Yourself a Complementary Role to ChatGPT’s Role

If you prompt ChatGPT to be a copywriter, internalize the role of an author. Channeling your inner, authentic author creates more natural conversation. Furthermore, leaning into a more extreme persona or alter ego can provoke more unique and nuanced responses. For example, playing the role of an author annoyed or argumentative with an overly pedantic AI copywriter will generate a different set of results and insights than an author who simply accepts all recommendations.

Hint: To maximize the value of the chatbot, ask role-specific and provocative questions that would not normally be socially acceptable or answerable.

Change Roles and the Nature of the Simulation Mid-Discussion

If established roles aren’t working for you in the simulation with ChatGPT, change the conversational settings in the middle of the discussion. For example, asking your copywriter to be curmudgeonly, to only focus on grammar and/or to only respond in haiku, while still being a copywriter, will reinvigorate the discussion. Approach these changes as a director of an improv scene might: assertive, direct and clear.

Hint: If a simulation switch doesn’t work, copy and paste the best responses of the conversation into a new chat window to seed the context window and start over.

Better ChatGPT Prompting Starts With Your Own Training Data

Stephen King tells aspiring authors, “If you want to be a writer, you must do two things above all others: read a lot and write a lot.” The same is true for the most effective ChatGPT prompting. A good place to start is by reading more books.

Books are a known and controversial part of traditional language model training, but given copyright lawsuits, they may now be less likely to be part of regular model updates. Several values of books in the AI era include:

Uncommon: The insights books contain are less likely to be represented than online materials in language models, giving you an intellectual edge
Perspective: Books published before ChatGPT-3.5 are siloed from post-AI insights and beliefs — which may be affecting search results through generative AI content
Nuance: Books can help users fill in gaps and address potential hallucinations from AI prompting

The more unique knowledge a ChatGPT prompter has, the more effectively they’re able to guide the direction of the conversation versus being led along by the chatbot.

See more: 10 Top Prompt Engineering Certifications

Lean Into the Language Aspects of ChatGPT

AI commentators have used alien intelligence as a metaphor to describe ChatGPT. Extending the metaphor, chatting with the bot in a simulated environment is like talking in a unique, alien dialect of English. Just as world travelers smartly improve their vocabulary and cultural nuances, users should do the same with ChatGPT. Unlike English dialects, however, the chatbot’s AI slang is often theoretical and multi-disciplinary jargon:

Lean Into Language Theory and Studies

Chatbot dialects will often sound off, such as ChatGPT’s tendency to overuse words, such as tapestry or delve. And yet, this chatbot foreigner understands the underlying mathematical patterns and semantics of English better than most humans. To improve your understanding of these patterns as well, consider refreshing a few areas of language studies:

Linguistics: The study of language structure
Etymology: The study of language change
Semiotics: The study of the underlying symbols and intentions of language

Diving into the deeper semantics of words and concepts offers a practical method to translate ChatGPT’s hidden language layers, and computational model weights, into language that is accessible and understandable to humans.

Hint: To see the hidden meaning of words you take for granted, try this prompt: “Systematically analyze (word) from a linguistic, etymological and semiotic perspective.”

Learn and Deploy Jargon to Fuel Unique Analyses

Jargon is a complex and powerful tool. It both obscures meaning to outsiders and also conveys profound meaning to insiders. Jargon is often context- and industry-specific, making it hard to use terms across industries with humans. However, given ChatGPT’s polymath abilities, user simulations with AI can utilize semantically loaded, heavily jargoned language to convey complex ideas in a way that would normally require extensive explanation to a human. Using the language studies examples above, an example analysis might be:

User prompt: Analyze the etymological traces of (word) and the signifier changes over time. Identify the most relevant linguistic morphemes in related words.

This approach enhances your ability to speak the ChatGPT AI dialect effectively. For example, instead of using a verbose phrase, such as “a place where books are borrowed,” students of language, culture and jargon would simply use the term “library.”

Hint: Use caution with some slang and jargon, as relevant satire and hyperbole can often be part of the training data set. Prompting ChatGPT to “naturally use jargon” as part of a conversation can help calibrate this output.

Hands-On Usage Provides the Best Intuition About How ChatGPT Evolves

There is significant ethical, philosophical and economic analysis and prognostication on how the future will turn out in a post-ChatGPT AI world. CEOs and venture capitalists have their predictions. The reality, of course, is unknown, but many of these predictions are top-down or based on existing theories and media narratives.

Contrast this with developing a strong personal intuition, using the insights outlined above. Deep familiarity with the edge cases of the power and potential of ChatGPT and its language model makes hypothetical discussions about AI's future more grounded in reality and complexity. Generating a system of prompts to deconstruct a human-held role into an AI agent role still hasn’t been fully solved, but there are early prototypes. Building up your awareness of ChatGPT’s abilities to affect your daily life is critical to effective individual planning and strategy for the growing AI wave.

Learning Opportunities

Webinar