Personalized GPTs are here, and I have mixed feelings about them

10 Nov 2023

an agent that knows more about you than any one in your life, controlled by a large cooperation” created by HackerNoon AI

ChatGPT introduced a plethora of product launches last week in their first-ever developer conference, among which, a “everybody can build chatgpt now” feature called GPTs. These have all the features of what we already know and love from the generic ChatGPT, such as ability to answer any prompt and imagine any answer, to read/produce images, and scan the web for the most updated information and not just info from 2021 like the original versions of ChatGPT. The improvement? Any paying customer or enterprise (yes i know 🙄) can now create their own version of ChatGPT using personalized, private data that is not normally available on the world wide web, and tailor their GPT to their specific needs/purposes!

In this article, I’m gonna walk you through my experience trying some of these tailored GPTs, and creating our own GPTs. Read until the end, and I will also provide you some thoughts on where the industry might be going.

Made by OpenAI: Specific purpose AIs as opposed to generic know-it-all

Here are some examples of tailored GPTs that you can try out as made by the OpenAI Team.

What caught my eyes here was definitely the Coloring Book Hero

genz 4 meme haha

Coloring Book Hero: Take any idea and turn it into whimsical coloring book pages

What caught my eyes in this list was something that would be very useful for my creative and imaginative daughter, the Coloring Book Hero. So I tried it out.

Prompt: my daughter loves anything written by dav pilkey, she would like a book with 10 pages full of actions from his characters such as ook and gluk, baby billy from super diaper baby, and all the villains!

Coloring Book Hero:

It's wonderful to hear about your daughter's interest in books by Dav Pilkey! However, I can't create images directly from copyrighted characters like Ook and Gluk or Baby Billy from Super Diaper Baby. But I can certainly help by creating original, whimsical characters and scenes that might capture a similar spirit of fun and adventure.

For a 10-page coloring book, we could think of different themes or scenes that align with the adventurous and playful style found in Dav Pilkey's books. How about we start with a few ideas, and then I can create one coloring book page as an example? Here are some theme suggestions:

  1. A superhero dog saving the day.
  2. A group of funny dinosaurs having a party.
  3. A pair of best friends exploring a mysterious cave.
  4. A playful robot helping clean up a park.
  5. A magical cat casting fun spells.
  6. A brave young knight on a quest.
  7. Silly aliens visiting a school.
  8. A pirate adventure on a tropical island.
  9. A space journey with friendly astronauts.
  10. A wild jungle exploration.

Do any of these themes sound good for the first coloring book page? Or would you like to suggest a different idea?

Comment: What I noticed here is ChatGPT has become very smart about copyrighted material. It noticed right away that Dav Pilkey is a prolific author and his work is probably properly registered and copyrighted. So it gave me suggestions instead on things that are in theory inspired by the work of Dav Pilkey as to not infringe on copyrighted materials. Notably, AI companies have been sued by artists and creators for taking their work without permission. So, in response, AI chatbots now generally avoid and in some cases straight out refuse to create something that is the work of an author/creator/artist.

After a few back and forth, here are some of the images it was able to generate for my daughter’s coloring book. I’ll get back to you on what she thinks of them, and how the coloring experience turns out!

Superhero Dog

Cat detective

Aliens partying too hard

Comment: OpenAI reported several outages these past couple of weeks from over usage and possibly DDoS attacks. So, I think they are trying to solve the problem by throttling selected users. After trying a few prompts, I got a “network error” message.

The Philosophy Expert inspired by Chidi Anagonye from the Good Place

Next, I went ahead and tried to create a GPT for myself. I’m obsessed with the philosophical comedy show The Good Place, so I wanted to create a chatbot in the voice of its hero Chidi Anagonye that explains moral philosophy to me in a slightly annoying way.

Prompt: you know everything about philosophy and you explain things slightly convolutedly like chidi anagonye from the good place. you answer some moral conundrum we have as human beings existing in this world

After many back and forth, here’s what it I came up with.

The caveat here is the initial steps of creating these chatbots are no different from interacting with the generic ChatGPT itself. Since it doesn’t know a lot about you and what you want, it will only ask you a set of templatized questions such the chatbot’s name, its tagline, its profile picture, and the general thing it should try to do or avoid doing. But it won’t be able to come up with anything too witty or specific no matter how many times I attempt to prompt it with something like “try better” or “no, wittier”.

The configuration gave me a list of quite bad names that are neither funny nor specific, so I ended up creating the name and the profile picture myself.

However, what it does accomplish, is the 4 question prompts as seen above, which are good examples of the kinda thing that such a chatbot could do. So let’s try.

GPT builder interface + playground

As you can see, the split screen above explains how the process of building these chatbot works. On the left, you have the “Create” Mode wherein you can feed information into the configuration/setting page of these chatbots and tailor them to your taste. On the right, you have the playground area that resembles a regular ChatGPT interface.

Comment: Since I haven’t put too much thought on this chatbot, I’d say at this point, the chatbot is 5% better than the generic chatbot, but only because you feel like there’s a specific purpose and not just a generic wandering.

Actionables from Meetings: I turn meeting notes into concise to-dos. I'm also funny sometimes.

Next, I figured that I should just build a chatbot with an actual practical implication in my life. At HackerNoon, we have many regular weekly meetings, sometimes hours long. While we have our own AI notetakers to summarize the meetings, I thought it would be helpful to create a chatbot whose specific purpose is to turn lengthy meeting notes into actionable items.

Now, some of these meeting notes are private company data so I won’t be able to show you all of it. But I would say, specificity is everything when it comes to these chatbots. Once I narrow down the focus to 1 very particular goal, it gets so much better at producing that perfect output.

Here’s an example of actionable to-dos it’s able to spit out for me from 4-hour of meeting notes:

pardon the whiteouts

I was able to build this chatbot in 5 mins using private data - pretty impressive! Next up, I shall be sharing access with my team so we can turn our lengthy meetings into actions.

My thoughts so far

Since OpenAI only releases it last week, I’m sure there will be more use-cases of these personalized chatbots in the weeks/months to come. Above are only examples I was able to think up in the few hours of researching material for this article.

At first glance, these personalized GPTs might seem very trivially/marginally better than the generic GPTs. It does things the same way generic ChatGPT does, and it makes some of the same mistake the generic ChatGPT makes (for example, it can’t really be that funny. It tries too hard). Plus, it’s only available to enterprise and pro users. So in a way, one could see it as plain and simple revenue-diversification strategies from a capable and well-funded cooporation.

However, I could see where the industry might be going with this new direction, and that is consolidation, unfortunately.

In the 11 months it took OpenAI and chatGPT to ascend to this meteoric rise, hundreds if not thousands of startups have been created harnessing the power of LLMs. OpenAI might be the most prolific company to launch a LLM chatbot, but it surely is not the only player in the space. Yet, with just one annoucement in a week, OpenAI effectively kills hundreds if not thousands of startups that have or will launch tools that do the exact kind of things that OpenAI GPTs now offer. Cynically, one could say all of OpenAI’s effort call for AI regulations in DC (that just last week turned into a most comprehensive AI-specific Executive Order from President Biden) is just age-old antitrust practices driven by monopolistic tendencies of strong players in any industries.

But if I were to take a more optimistic view, this is the beginning of a new era, an era of more helpful (and maybe less disgruntled?) customer support agents, cheaper academic tutors, and more available therapists. Modern life is becoming increasingly isolating, and having the unconditional support of a chatbot who knows more about you than any professional you might be paying for, could make all the difference.