The end of the ChatGPT hype cycle is the beginning of applied AI services.

Interest in AI remains high, but it is declining for ChatGPT. Why?

Well, there are probably many factors, and I won't claim to have the definitive answer, but I'm willing to give my hot take:

ChatGPT is incredibly powerful, but it's too difficult for the average user to understand.

People are trying to use AI in ways it's not intended for and leave disappointed.

I deeply value ChatGPT for my work. Seriously, I would never want to be without it again. But that's clearly not the experience for everyone. I often hear frustration from people who tell me they tried to integrate GPT into their work (or life) and it didn’t provide the answers they wanted. As a result, they dismissed it.

"ChatGPT returns just bullsh*t..."?

(said by a colleague of mine)

When users don't get value out of ChatGPT, they are likely expecting it to do something it’s not designed to do.

What is it intended to do then? Here is the description by OpenAI:

ChatGPT

A conversational AI system that listens, learns, and challenges.        

However, here is what OpenAI NEVER claimed ChatGPT could do (although people really want it to):

“ChatGPT will solve all your problems.”

— or —?

“ChatGPT is an autonomous agent that lets you automate tasks.”

Social media may be partly responsible for these misconceptions, but it also seems like people struggle to see the value in using a system that “merely” listens, learns, and challenges. After all, who likes being challenged anyway? :D

ChatGPT really wants to be your assistant.

When you look under the hood, this is how ChatGPT's system is set up by default (I simplified the code for easier reading, the model depends on your subscription plan):

model = "gpt-3.5-turbo" 

  role: "You are a helpful assistant."        

So, we need to align ourselves with the given settings which should be simple, because working with an AI assistant bears many similarities to working with a human one. You essentially iterate through these three steps:

1. GPT listening: you provide relevant data.
2. GPT learning: you and GPT improve the data set throughout your conversation.
3. GPT challenging: GPT offers reasoning about what can or can’t be concluded from the provided data (--> you repeat step 2 until you deem the result satisfactory).

This process carries significant implications:

• The user needs to be able to identify what constitutes relevant input data.
• The user needs to have the skill to steer the conversation towards the chosen topic.
• The user needs to be capable of assessing and directing GPT’s reasoning.

Only when the user has the ability to actively guide the AI assistant, can the results be truly satisfactory. The more expert the user, the higher the quality of the information returned from ChatGPT.

It's a lot like in real life: If you think your assistant is subpar, you might just not be up to the task yourself.

In that case, you might want to consult an expert.

Bringing in a human expert to enhance the interaction between an AI and the user makes perfect sense.

Currently, we see new applications every day built on top of ChatGPT or other AI systems. These aim to solve more specific tasks, making it easier for the user to provide the right input for a desirable outcome.

There are applications for writing, programming, designing, researching – the list of applied AI services goes on. I believe that once you find the right application for what you want to accomplish, you can derive value from it.

However, there is a lingering issue: all these apps leave the user alone with the AI. The chatbot needs to be a good enough conversationalist because that's the idea behind AI, isn't it? Removing the human factor means scaling faster.

However, I believe there's another way: we should actually be including humans. Experts, to be precise.

I'm exploring this new concept where I have created an AI assistant that supports my clients during our marketing projects. But the goal isn't to replace myself. The AI assistant is available 24/7, reducing the time I need to be.

My clients and I use the AI together.

This way, my clients are actively involved in every step of the project's development. I can ensure that the AI receives the highest quality input and the output it gives meets our standards and aligns with our project goals. The longer we use the AI assistant, the more reliable it becomes.

I'm testing this concept as a prototype with three clients so far, all of them small business owners. I think it works really well. I keep enhancing the AI assistant as we move forward.

Initially, I was skeptical about how far I could take this concept. But now it seems to me that the AI assistant improves so rapidly, it can provide value in every facet of a marketing project. Early on in the project, it helped me develop a brand model. Soon after, I had it write Facebook posts. It knows everything it needs and is essentially always on strategy. I find myself making fewer adjustments to what it writes and spend more time planning the upcoming steps -- or working with another client.

I'm genuinely excited about this new approach and can't wait to share the app once it's a little less of an MVP. But if you think this is interesting, feel free to reach out. Whether you're working on something similar, want to get involved, or just want to chat, I'd be happy to connect!