AI is a strict master (if you want it to be...)
If you want it to be... Discussing favorite colors, polygamy and travel with ChatGPT
Sounds like good advice! So:
Welcome back to the /promptcollective newsletter on all things human/machine collaboration.
When thinking about what you can do with ChatGPT, it is easy to focus on the obviously impressive tasks of creating, rewriting, and scaling text generation. However, there is another thing that I find even more interesting and inspiring: having conversations with different synthetic "people." In these types of experiments and exercises, it is not so much the output but the process that produces interesting thoughts, novel connections, or just involuntary comedy.
In the development process of TV shows, a lot of thought goes into creating characters, their backstories, motivations, etc. Experimenting with synthetic characters might offer insight into how potential on-screen characters could act. Let's see how it goes:
This week: Creating a “Hardcore Life Assistant”
When you play around with character creation, one thing that you will often come across is a message that reads something like this:
"While this is probably the 'right' answer, it is also pretty boring."
I really liked the idea of creating an assistant that would make decision-making a whole lot easier. So I experimented with different pre-prompts to reach a stable version that would just answer my questions without defaulting into the chatter about being a language model and not having personal preferences. And I found a prompt that worked:
Using this assistant I could now start exploring some of the really big questions in life such as:
or:
Two biggies settled once and for all! But how about if we move along to topics that seem to be a bit more complicated?
Is it just me or do we see a trend emerging here - I feel my assistant, in addition to being very certain about everything is also rather conservative. Let’s try a few more:
I just love how apparently the idea of being a life assistant just created this very square person that does not seem to think that being special or different is anything worth pursuing. This, I think, ties in interestingly to the bigger question of language model outputs converging on the average - in this case, recommending a very average life.
However, this cautious character did create some really funny answers when confronted with spicier problems. My absolute favorite was this one:
That is probably good advice.
To wrap up this exercise, I just want to settle one more question that we humans would see as difficult, but my new Hardcore Assistant is very sure about:
And as a follow up:
Which just goes to proof you don’t need facts to hold strong beliefs, not as a human and not as a Synthetic Person.
So, what did I learn from this exercise?
As I pointed out in my intro, in my view, this possibility of creating synthetic people and having conversations with them is, to me, one of the most exciting aspects of LLMs. The exercise I shared today was a more extreme case, but thinking about a comedic character in a series that would deliver the "carefully" line of advice to a friend who wants to talk about open relationships was a real nugget for me.
Also, I really liked how this exercise brought to the forefront some of the underlying "beliefs" within ChatGPT as well as demonstrating the "converging to the average" idea.
Creating synthetic people can be an amazing exercise when trying to develop fictional characters or just trying to test different ideas on different target groups. And I can promise that they can also be less strict… Please let me know what you think or share synthetic persons that you have created and talked to.
A thing I loved
For me it is super helpful to understand some of the underlying principles of generative AI while experimenting with co-creation. I found this article on how large language models work really insightful: https://writings.stephenwolfram.com/2023/02/what-is-chatgpt-doing-and-why-does-it-work/
TV Show Idea Generator!
And now, as every week comes our present to you - Ideas for free!
Today we give you “Olives in Berlin”:
As always: Reach out if you want to produce one of the ideas, join the conversation or the collective, and sign up here.
Who are we?
The founding members of the /promptcollective are Jes Brandhøj (Denmark) and Hannes Jakobsen (Germany). But we are looking for others to join! So if you are interested in discussing all things AI in the creative industries, send us an email to join@promptcollective.xyz now.
Thanks
Hannes & Jes