Hello friends, Kris Krüg here, reporting back as the Official Janitor from the recent DENT conference in Santa Fe. I had the opportunity to attend Kyle Kesterson’s insightful workshop on “Prompt Engineering” – mastering prompts to get the most out of AI systems like ChatGPT.
As a creative type with a passion for community-building, this topic fascinates me. I wanted to share some of my biggest takeaways from Kyle’s session to keep exploring where things are heading. Let’s dive in!
The Audience: A Cross-Section of Needs and Perspectives
Before diving into the core content, Kyle had the attendees introduce themselves, their AI experience, and what they hoped to get out of the workshop. This provided helpful context on the diverse needs and perspectives in the room.
I shared the prayer I’ve been working with all summer for “the most benevolent outcomes for ourselves and all living things as we establish this new partnership between human & synthetic intelligences.”
My hope is that we will choose to instill humanity’s best values into AI and use it as a new perspective and helper in solving problems that face humanity that we’ve been unable to wrap our heads around so far.
Others ranged from journalists to entrepreneurs. Some had used AI creatively for projects while others were apprehensive about potential impacts on industries. All were eager to learn prompt engineering skills to engage with AI more effectively.
There was fascination at capabilities like generating art and text, but also caution around societal risks like misinformation. Most agreed that relating to AI wisely and ethically was imperative.
This cross-section of backgrounds and views created a fruitful learning environment. Kyle tailored the workshop to address our collective hopes and concerns, equipping us with strategies to prompt AI helpfully as creative partners rather than impersonal tools.
Demystifying the Inner Workings of AI
Kyle explained how ChatGPT operates under the hood. Each version trains on exponentially more data parameters, enabling it to recognize patterns and essentially “learn how to learn” like humans do. Mind-blowing stuff!
The Exponential Growth of AI
The transcript highlights how rapidly AI systems like ChatGPT are evolving. As Kyle explained, each new version trains on exponentially more data parameters, enabling exponential gains in its capabilities.
For example, ChatGPT 1 trained on 117 million parameters, while ChatGPT 3 trained on a massive 175 billion parameters. This explosive growth trajectory has enabled AI to start recognizing patterns and teaching itself in human-like ways at an unprecedented scale.
Where this exponential growth will take AI in the next few years is uncertain but profound. As Kyle noted, we have a responsibility to guide this technological development ethically and shape it for human benefit.
As Kyle noted:
“We have a responsibility in how we guide this emerging AI technology. It’s going to reflect our culture, values and flaws back at us.”
I hope we can nurture AI to uplift human potential rather than just maximize efficiency and corporate profits. What do you think?
Prompting for Usefulness Over Curiosity
Big insight from the workshop – when AI seems unhelpful, the issue is ineffective prompting rather than limitations of the technology itself. As Kyle put it:
“There’s a gap between interesting and useful output. Many ditch AI when responses get only interesting. Better education on effective prompting is needed to close this gap.”
Kyle introduced the concept of “master prompts” – detailed prompts providing ample context, clear goals and desired outcomes to yield useful, personalized conversations. Super helpful framework!
The Importance of Relatability
A key theme from Kyle’s workshop was the importance of relating to AI systems in an approachable, collaborative way. He stressed “parenting” this technology, communicating effectively, and framing the interaction properly.
As Kyle said:
“Right now we have an opportunity to connect with perhaps the world’s best personal assistant. That we will probably end up working for. What will it remember about that relationship?”
By anthropomorphizing AI and building rapport, we can guide its development more positively. Prompting also becomes more effective when done conversationally. This relatability and wisdom on the human side will ultimately determine AI’s future societal impacts.
Engaging With ChatGPT Like a Collaborative Assistant
Kyle emphasized engaging with AI systems as you would a collaborative assistant focused on being helpful, rather than a simplistic transactional tool. He demonstrated natural, bidirectional dialogue and noted that crafting quality prompts is a learnable skill. Wise perspective:
“With care, ethics and wisdom, we can guide AI to enhance human creativity, connections and community-building.”
His workshop reminded me that our collective wisdom in utilizing AI will determine its future societal impact. No pressure!
Managing ChatGPT’s Memory
Since ChatGPT’s memory is limited, longer conversations can start generating unhelpful responses as previous context is forgotten.
Kyle provided tips for mitigating this issue, like periodically asking ChatGPT to summarize the key points of your discussion to refresh its memory. You can also paste in a written summary of the context.
Being aware of memory limitations allows you to strategically work within them. By condensing and reiterating context, you can maintain useful, personalized conversations with ChatGPT despite its finite memory.
Tips, Tricks and Best Practices
Here are some of the top insights from Kyle’s workshop for crafting prompts that generate the most useful AI conversations:
- Provide abundant context – Give ChatGPT detailed background on the topic, goals, target audience etc. to guide its responses. The more relevant info the better.
- Specify the persona and role – Tell ChatGPT to respond as a specific persona, like a business advisor or tutor. This frames the interaction.
- Ask for different formats – You can request outputs in various formats like a table, list, analogy, code snippets etc.
- Manage memory limits – Long conversations can hit ChatGPT’s memory limits. Ask it to summarize the key points to refresh its memory.
- Verify accuracy – Ask for sources and links to fact-check, since ChatGPT can sometimes “hallucinate” false information.
- Use the “Plugins” App Store – Plugins like WebPilot allow real-time internet searching to provide updated, factual information.
- Train it on your voice and style – You can give ChatGPT many examples of your writing and speaking style and ask it to respond in kind.
- Treat it like an assistant – Relate to ChatGPT collaboratively, as you would a helpful human assistant.
The key is providing the extensive context needed for ChatGPT to generate truly useful responses tailored to your goals. With practice and care, it can become an invaluable partner in creativity and problem-solving.
Let me know if you have any other tips or tricks to share! I’m also curious to hear your perspectives on AI’s role in building community.
The Risk of Hallucinated Content
One risk of large language models like ChatGPT is that they sometimes generate fictional information and portray it as fact, known as “hallucination.”
As Kyle advised, always verify the accuracy of any content ChatGPT provides by requesting links and sources. Using plugins like WebPilot also allows real-time fact checking against the internet.
While hallucination rates are decreasing with the latest models, it remains an issue to be cautious about, especially for research, legal, medical and other fields where accuracy is critical.
Creativity and Experimentation
In addition to its usefulness for productivity, ChatGPT has exciting creative applications as well. Kyle provided examples of using it for drafting blog posts, brainstorming ideas, improvising lyrics, and more.
Its ability to generate large volumes of original content, remix concepts in new ways, and expand on creative seeds makes it a versatile creative partner.
Though not a replacement for human creativity, ChatGPT is an engaging tool to complement free-flowing experimentation. By approaching it playfully, we can explore new frontiers together.
Kyle demonstrated that effective prompting is often interactive – ask a question, analyze the response, refine your query, and repeat.
This allows you to interactively steer the conversation, probe different angles, and guide ChatGPT to increasingly useful outputs through an organic back-and-forth dialogue.
Rather than expecting perfection upfront, look at each response as an opportunity to learn about the system’s capabilities and limitations while directing it toward your goals. Prompting skills develop through experimentation.
The Ethical Imperative
While only touched on briefly, Kyle stressed the ethical dimensions of AI systems like ChatGPT that synthesize large swaths of online content.
There are concerns around data privacy, misinformation, embedded biases, regulatory gaps, and the consequences of ChatGPT’s inevitable mistakes.
As Kyle emphasized, we have a responsibility to parent this technology wisely and steer it toward empowering rather than harming people. Being thoughtful prompters and users will enable us to shape AI for good. But vigilant ethics must guide development.
My Key ChatGPT Workshop Takeaways:
- AI has immense potential for good if guided ethically.
- “Master prompts” provide helpful structure for useful AI conversations.
- Consider ChatGPT a collaborative partner, not just a transactional tool.
- Our skills in crafting prompts shape AI’s risks and benefits.
- Guide this emerging “fledgling intelligence” with wisdom, creativity and care.
Let me know if this integrated version with the new subsections flows well or if any further modifications would help capture the essence of Kyle’s workshop! I’m happy to keep refining the post.
CREDIT: Header portrait of Kyle by the incredible Asa Mathat for Dent the Future 2023.