Whatever the disadvantages may be, I found ChatGPT can help me boost my creativity when I feel stuck.
But before I start I have to make a confession. I can’t write anything about ChatGPT without mentioning that as a conversation designer I can only swoon at ChatGPT’s level of complex intent recognition. It ‘only’ takes about 750 GB of data to start with :-).
Teaching machines to ‘understand’ human language is complicated
People who don’t know anything about conversation design might be completely unaware of this, but getting chatbots to recognize even 1 short, basic question or remark – often called ‘intent’ – is a lot of work. You’ll have to feed the platform a lot of training data in the form of sentences. And those have to be very precise.
What’s natural for you and me isn’t natural for chatbots
From the training sentences the machine can distil recurring patterns and based on these patterns it will make predictions when it encounters other sentences: things users say while chatting with the bot. If the machine reaches a predetermined level of confidence that the user statement belongs to a certain group, it will show the chatbot response that is matched with that specific group.
Recognizing and responding to intents
For example, you want your chatbot to recognize the intent “greeting” if a user types a message like “Hello!” or “Hi there!”. And to give an answer that you linked to that intent, like “Hello! What can I do for you today?”.
The more intents you add, however, the greater the chance of confusion. Every additional training sentence can impact the recognition of all the other training sentences. And I’m not even talking about subtle differences that may be caused by different contexts yet – differences that are easily picked up by humans, but not so much by machines.
LLM’s as a tool for conversation designers
Maaike Groenewege already explored some ways conversation designers could profit from using LLM’s to help them with tasks that are very time-consuming and mean a lot of handwork for most conversation designers in most bot platforms.
But I’m digressing from my story. As eager as I am about exploring that as well, I actually want to highlight something else here.
ChatGPT as a catalyst for your creativity
This week I wanted to write a post on LinkedIn about the agency I work for, but I had no inspiration. So I decided to go back to my first love – theater – and do another little experiment with ChatGPT.
I instructed ChatGPT to write a theatrical dialogue in Shakespeare’s style, in which one character tried to convince a second character why it would be a good idea to apply for a job at Crossphase. I also gave 3 reasons that I wanted to incorporate in the text.
Behind the scenes
I set the frame and ChatGPT made up a dialogue.
I told ChatGPT to use real names instead of ‘Character 1’, ‘Character 2’.
So it made up a new dialogue.
I told it not to use ‘Hamlet’ and ‘Ophelia’.
Why? Well, I somehow think it’s better for potential colleagues not to start associating Crossphase with tragic heroes or heroines who often endure and cause great suffering before meeting untimely deaths.
We went back and forth – I asked for a title as well, suggested shorter ones, and did some tweaking at the end.
A little bit of me, a little bit of ChatGPT and a lot of Shakespeare
IMS: Instead of ‘Personage’ or ‘Character’, can you use real names, like Shakespeare could have given to the characters in his plays?
ChatGPT: Oh, my apologies. Here is a revised version with appropriate character names
IMS: Uhm – and can you please use names that are similar to names Shakespeare used for his characters, but that he never chose?
ChatGPT: Very well, here is one more revised version with appropriate character names
After a few iterations, this is the end result:
Portia: Good morrow, Benedict! Pray tell, why dost thou hesitate to join the ranks of Crossphase?
Benedict: Oh, I know not. I fear I may not be up to the task.
Portia: Fear not, dear friend. Crossphase is a place of growth – both professional and personal. I, myself, have developed my own training program, and there are others who have crafted courses for those who aspire to be web editors or conversation designers.
Benedict: That does sound enticing. But what of our colleagues? Are they not fierce competitors?
Portia: Nay, we have a young and enthusiastic team, full of eager learners. And fear not, for Crossphase also hosts delightful activities outside of work to foster camaraderie among us. I think you would really thrive at Crossphase.
Benedict: Well, that does ease my mind. Perhaps I shall consider joining Crossphase after all.
Portia: Excellent! Thou wilt not regret it, I assure thee
ChatGPT: cause for concern?
If you believe that ChatGPT is without fault, its answers are always reliable, and you’re unwilling or unable to reflect critically on anything it says, then I think that’s worrying indeed.
However, I do believe we can benefit from ChatGPT without unquestioningly accepting or reproducing everything it states.
No, it doesn’t deliver perfect articles I can use without editing right away. But when I see what it produces, it becomes clearer to me exactly what I’m looking for. This means I can fine-tune my question or assignment accordingly – and ChatGPT adjusts quickly when I do.
I like the way it helps me sharpen and speed up my thoughts – a bit like the effect brainstorming out loud can have. Like I said, it can help boost your creativity. And because ChatGPT generates text so quickly, you can quickly advance a few steps.
So – what do you think?
When I wouldn’t have chosen to play ping pong with ChatGPT, I probably wouldn’t have thought of trying to capture my positive experiences in a theatrical dialogue either. In that case you would have probably gotten a more straightforward text.
While there’s nothing wrong with straightforward texts, I like the way this turned out. It felt more like a fun pastime than a self-imposed exercise that I got little pleasure from – even though I really mean what I say.
I’m curious what you prefer?
Want to read more on ChatGPT?
- By OpenAI, the company that trained the model called ChatGPT, 30 November 2022:
ChatGPT: Optimizing Language Models for Dialogue
- By Alberto Romero (analyst at CambrianAI), 2 December 2022 :
ChatGPT Is the World’s Best Chatbot
- By Maaike Groenewege (conversation designer, linguist & coach), 4 – 11 December 2022:An evening with ChatGPT – Part 3: Hey ChatGPT, help me with NLP!
- By Andrew from DeepLearning.AI – The Batch, 7 December 2022 :
When Models are Confident — and Wrong – Language models like ChatGPT need a way to express degrees of confidence.
- By Thomas Smith (among other things SEO expert, journalist), 8 December 2022:
ChatGPT Is Having a Thomas Edison Moment – Why breakthrough technologies need to be accessible
- By Abeba Birhane (PhD, senior fellow in Trustworthy AI at Mozilla Foundation) and Deborah Raji (computer science PhD student at UC Berkeley and Trustworthy AI fellow at Mozilla Foundation), 9 December 2022:
ChatGPT, Galactica, and the Progress Trap
- By Maximilian Schreiner (managing editor at THE DECODER. As a trained philosopher, he deals with consciousness, AI, and the question of whether machines can really think or just pretend to.), 24 December 2022:
ChatGPT review: Does OpenAI’s text AI beat our editorial team?
or in German:
ChatGPT im Test: Schlägt OpenAIs Text-KI unsere Redaktion?
- By Gary Marcus (scientist, best-selling author, and entrepreneur), 25 December 2022:
What to Expect When You’re Expecting … GPT-4 – What comes after ChatGPT? 7 predictions for 2023