ChatGPT Enables Costless Creativity

Get your creative juices flowing by mastering the AI

One interesting thing about the rise of generative AI tools like ChatGPT is that the cost of idea generation has declined to zero. And when costs decline to zero, interesting things start to happen. Let’s think a bit about what it means now that creativity is nearly costless.

In principle, all it takes is some good prompt engineering to get ChatGPT or Dall-E or Stable Diffusion to create something magical. Here’s Nick Bilton writing in Vanity Fair:

Members of the pro-AI tech set concede that this technology has the potential to automate many tasks that today require human creativity, but they point out that machines are not truly capable of understanding or appreciating art in the same way that humans are. Machines do not have consciousness; a computer can’t feel what it’s like to fall in love or lose a loved one or be tormented to a point that you have to chop off your ear. The argument goes: Machines can mimic our creations, but they cannot truly understand the emotions and experiences that inspire us to create. But, to me, if machines are able to imitate art with emotion and depth because they are learning from things humans have created over hundreds of years, then the machines are, in turn, an extension of those human emotions. A machine does not have to be conscious or capable of experiencing emotions to create art that is meaningful to us. The value and significance of the art lie not in the machine’s ability to feel, but in the ability of the viewer to appreciate it.

Let’s set aside the philosophical question of whether machines can ever be conscious, or whether they are an extension of human consciousness. While those ideas are certainly interesting, they merit their own examination which is beyond the scope of what I’m contemplating here. As I’ve written before, the best way that I have come up with to understand ChatGPT and other generative AI tools is to think of them as a complement for your cognition. Or an adjunct for your brain. I call ChatGPT an infinitely scalable second brain here. The bottom line is that ChatGPT and related tools allow people to generate ideas at scales heretofore unknown. And that is worth exploring further.

Virginia Postrel notes that zero-cost creativity allows for creative abundance:

AI-driven “sketching” allows a one-person shop to take on more ambitious projects. It lets creators go quickly from idea to concept art. It makes it cheap and easy to try out ideas that would be too costly and time-consuming to experiment with using human labor. At this point, the results are too crude to appear on the screen but they give human artists a place to start working.

The source of all this creative output is prompt engineering. A well-structured prompt is an engineered prompt, and it is meant to elicit from the AI output that is relevant to the person’s inquiry. When I tried to get ChatGPT to build a math tutor in Python, which would help a student understand matrix multiplication, addition, and subtraction, I had trouble getting ChatGPT to produce the output I expected. I had to engineer my prompts. Refining my prompt required me to think in a more structured way.

Here’s Jon Evans exploring what’s possible with robust prompt engineering:

For whatever reason—we are a visual species living in a visual world; generative art is far more controversial than generative language; AI art’s effort-rewarded feedback loop is tighter and more visceral; its prompts tend to be somewhat less complex, chained and dialectical—there are far more guides to DALL-E and Stable Diffusion prompting that guides to large language models.

The embarrassment of riches includes, first, an excellent overview of the state of AI art, including a thoughtful analysis of legality and morality, at “AI Art Panic.” But if you want to just get to the prompts, go straight to two really excellent pages of examples: you can learn a lot just from scanning through “Best 100+ Stable Diffusion Prompts” and “Top 50 Prompts for Midjourney and DALL-E.”

Where am I going with all of this? Think about very creative people such as Stephen King. They are very creative because they never stop generating ideas. In some sense this is tautological: Stephen King is creative because he generates a lot of ideas. And yet, it points to a certain truth about generative AI: generative AI allows you to be much more creative because it allows you to create many more ideas.

But where ideas used to be very expensive, they are no longer. They’re free or nearly so. As the world has gotten richer the cost for creating any given idea has declined. And we’re in a place where the cost to come up with some ideas is near zero. That has profound consequences for people, for society, for technology, and for progress.

The pessimist will read that paragraph and think “Great! More junk to wade through.” And this is true! Infinitely scalable idea creation means that much more junk will be created. Curation will become increasingly important. A prompt engineer who tells ChatGPT to come up with 25 ideas to market a new widget will have to curate that morass of ideas and extract the gem. A prompt engineer who tells DALL-E to generate 25 logos for her client will likewise have to curate its output. And there’s no reason, for either of these prompt engineers, to keep the output to 25 ideas. Why not 250 or 2500 ideas? At some point, of course, for any given purpose, the number of ideas approaches diminishing marginal returns. And the prompt engineer will have to figure out what that number is.

But in principle, nothing is stopping our prompt engineer from telling ChatGPT to generate a million ideas.

So, we’ve entered a world of idea abundance. Much of knowledge work—what we commonly call white collar work—depends fairly heavily on robust idea generation. But how do you generate those ideas? Brainstorming with your colleagues is one way. Iterating on prompts—prompt engineering—is another. Here’s Wharton professor Ethan Mollick writing about this:

A way to generate good ideas is to change the boundaries of the problem by altering the constraints you face. This allows you to see a problem in a new way, which might lead to new solutions. Adding constraints can have the same effects.

Psychological experiments have shown that when people have limitations on the ideas they can pursue they are often, paradoxically, more creative. For example, in one study, participants were told to design a toy for children, given five weirdly shaped blocks selected from a larger set of pieces. In some cases, the experimenters randomly gave people blocks, while in others they were allowed to pick their own. Additionally, in some cases, the subjects had to use all five blocks, or they were allowed to use any numbers of the blocks they were given. The least constrained scenario, in this case, allows for someone to pick their own blocks and use whatever amount they see fit. Surprisingly, people in this group came up with the least creative answers. Instead, the most constrained group, where people were given five blocks randomly and had to use all five, was actually the most creative, both in process and outcomes. Constraints help because they force you to change your thinking and break out of old frames. You can add constraints to your ideas in various ways, often by consciously changing your perspective.

What does this look like in practice? Following is ChatGPT’s response to the prompt Give me 10 ideas for blog posts about how to use you for market research.

1. "5 Ways to Use Assistant for Market Research: A Beginner's Guide"
2. "Maximizing the Value of Assistant for Market Research: Tips and Tricks"
3. "Conducting Primary Research with Assistant: An Overview"
4. "Using Assistant to Analyze Secondary Research: Best Practices"
5. "How Assistant Can Help You Stay Up-to-Date on Market Trends"
6. "Using Assistant to Identify and Analyze Your Target Market"
7. "Assistant's Role in Competitive Analysis for Market Research"
8. "Leveraging Assistant for Customer Segmentation in Market Research"
9. "Using Assistant to Generate and Test Marketing Hypotheses"
10. "Assistant's Role in Evaluating the Effectiveness of Marketing Strategies"

I asked it to write a 500 word blog post about “5 ways to use assistant for market research: a beginner’s guide.”¹ Here’s its output:

As a large language model trained by OpenAI, Assistant is a powerful tool for conducting market research. Whether you're a business owner, marketer, or researcher, Assistant can help you gather and analyze data to inform your decisions and strategies.

Here are five ways you can use Assistant for market research as a beginner:

1. Identify and analyze your target market: One of the first steps in any market research project is to identify your target audience. Assistant can help you gather data on demographics, interests, and behaviors of potential customers. You can also use Assistant to analyze this data and identify trends or patterns that can inform your marketing efforts.

2. Conduct primary research: Assistant can help you design and administer surveys or focus groups to gather primary data from your target market. You can use Assistant to create a survey questionnaire, distribute it to your target audience, and analyze the results. This can provide valuable insights into customer needs and preferences.

3. Analyze secondary research: Assistant can also help you find and analyze secondary data sources, such as industry reports, news articles, and government statistics. This can provide a broad overview of the market and help you understand trends and patterns. You can use Assistant to search for relevant data sources, extract key information, and summarize it in a report or presentation.

4. Stay up-to-date on market trends: Assistant can help you stay informed about the latest developments in your industry or market. You can use Assistant to set up alerts for specific keywords or topics, and receive regular updates on news and trends. This can help you stay ahead of the curve and make informed decisions.

5. Evaluate the effectiveness of marketing strategies: Assistant can also help you evaluate the effectiveness of your marketing efforts. You can use Assistant to analyze data on customer engagement, conversion rates, and other metrics to understand how your strategies are performing. This can help you identify areas for improvement and optimize your marketing efforts.

In conclusion, Assistant is a versatile and powerful tool for conducting market research. Whether you're a beginner or an experienced researcher, Assistant can help you gather and analyze data, stay up-to-date on market trends, and evaluate the effectiveness of your marketing strategies.

This writing isn’t especially interesting, but it’s functional. And it creates a lot of possibility. I could post the blog post somewhere, I suppose. Or I could take the blog post’s fifth suggestion, “evaluate the effectiveness of marketing strategies” and use that as a jumping off point for more inquiries: “I want to collect data on customer engagement and conversion rates. What’s the best way to do this?”

Here is ChatGPT’s output:

There are several ways you can collect data on customer engagement and conversion rates to evaluate the effectiveness of your marketing strategies. Here are a few options:

Web analytics tools: You can use web analytics tools, such as Google Analytics, to track customer engagement and conversion rates on your website. These tools can provide insights into website traffic, bounce rates, and customer behavior.

Surveys and focus groups: You can use surveys and focus groups to gather customer feedback and understand their experiences with your brand. This can provide valuable insights into customer engagement and conversion rates.

Social media analytics: If you have a presence on social media, you can use social media analytics tools to track engagement and conversions. These tools can provide insights into the performance of your social media campaigns and help you understand which platforms and strategies are most effective.

A/B testing: A/B testing is a method of comparing two versions of a marketing campaign or website to determine which one performs better. You can use A/B testing to compare different versions of marketing materials or website elements and measure the impact on customer engagement and conversion rates.

Customer relationship management (CRM) software: If you have a customer database, you can use customer relationship management (CRM) software to track customer interactions and conversions. This can provide valuable insights into customer engagement and help you identify trends or patterns that can inform your marketing strategies.

Ultimately, the best approach will depend on your specific goals and resources. You may want to use a combination of these methods to gather a comprehensive view of customer engagement and conversion rates.

If I knew nothing about customer engagement and conversion rate analysis before, I know a lot now!

What if I want to automate some social media analysis? Well, here's my prompt: Can you give me an example of a python script that I could use for doing social media analysis?²

Here’s its output:

import tweepy

# Enter your Twitter API credentials
consumer_key = "YOUR_CONSUMER_KEY"
consumer_secret = "YOUR_CONSUMER_SECRET"
access_token = "YOUR_ACCESS_TOKEN"
access_token_secret = "YOUR_ACCESS_TOKEN_SECRET"

# Authenticate with Twitter's API
auth = tweepy.OAuth1UserHandler(
    consumer_key,
    consumer_secret,
    access_token,
    access_token_secret
)
api = tweepy.API(auth)

# Set the hashtag you want to search for
hashtag = "#marketresearch"

# Collect data on the hashtag
tweets = tweepy.Cursor(api.search_tweets, q=hashtag, lang="en", tweet_mode="extended").items(100)

# Iterate through the tweets and print out the text
for tweet in tweets:
    print(tweet.full_text)

It includes this explanatory text:

Here is an example of a Python script that you could use to collect data from a social media platform using its API (Application Programming Interface). This script uses the Python library "tweepy" to access Twitter's API and collect data on a specific hashtag:

This script will collect 100 tweets that contain the hashtag "#marketresearch" and print out the full text of each tweet. You can then use this data for further analysis or to identify trends or patterns related to market research on Twitter.

Keep in mind that you will need to obtain a set of API credentials from Twitter in order to use this script. You can find more information about how to do this in the Twitter API documentation.

I hope this example is helpful! Let me know if you have any questions.

So let’s review:

• I couldn’t come up with new ideas for Substack posts about what I could use ChatGPT for.

• I recalled Ethan Mollick’s Substack post about unlocking creativity by outsourcing the brainstorming to ChatGPT.

• So I used ChatGPT to come up with 10 ideas for blog posts about market research.

• I asked ChatGPT to write a blog post about “5 ways to use assistant for market research: a beginner’s guide.”

• Prompted by that post I asked ChatGPT to write a simple script that I could use to automate social media analysis, and it gave me one that I can use to parse Twitter for certain keywords.

And the larger and more important point is this: you can repeat this process for almost any idea or concept you can come up with. Want ideas for cupcake flavors? Here they are:

1.Salted caramel: A classic combination of sweet and savory, salted caramel cupcakes are made with caramel-flavored cake and topped with a caramel buttercream frosting.

2.Lemon lavender: These refreshing cupcakes are made with a lemon-flavored cake and topped with a lavender-infused buttercream frosting.

3.Peanut butter and jelly: A childhood favorite, peanut butter and jelly cupcakes are made with a peanut butter-flavored cake and topped with a raspberry jam filling and a peanut butter frosting.

4.Red velvet: A classic red velvet cupcake is made with a rich, moist cocoa-flavored cake and topped with a creamy cream cheese frosting.

5.Apple spice: These autumnal cupcakes are made with a spiced apple cake and topped with a maple cream cheese frosting.

6.Chocolate hazelnut: A twist on the classic chocolate cupcake, chocolate hazelnut cupcakes are made with a chocolate cake flavored with hazelnut liqueur and topped with a hazelnut buttercream frosting.

7.Orange cream: These citrusy cupcakes are made with an orange-flavored cake and topped with a whipped vanilla cream frosting.

8.Gingerbread: A holiday favorite, gingerbread cupcakes are made with a spiced gingerbread cake and topped with a cream cheese frosting.

9.Mint chocolate chip: These refreshing cupcakes are made with a mint-flavored chocolate cake and topped with a chocolate frosting and chocolate chips.

10.Maple bacon: A unique and indulgent flavor, maple bacon cupcakes are made with a maple-flavored cake and topped with a maple frosting and crispy bacon bits.

The world’s your oyster, and ChatGPT is your substrate.

1. As a side note: it is interesting to me that ChatGPT has taken to calling itself “Assistant”!

2. Curiously, this prompt generates a warning message suggesting that I may be violating OpenAI’s content policy. A cursory review of its content policy doesn’t really explain why this message was triggered.