The Real Value of Custom GPTs

It's been a couple of weeks now since OpenAI introduced their new custom GPT feature for premium members. As a tech junkie, I've been relentlessly playing with them, developing numerous different applications and concepts.

One thing, to this point, that I can't get past is the low barrier to entry in this field, which is essentially the ability to type and a $20 bill for the ChatGPT upgrade. Other than that, it's just as OpenAI says - anyone can build their own GPT.

That isn't to say the technology isn't fascinating or of immense value. But, in these early stages, there isn't an obvious marketable value (although I have been around the tech industry long enough to know that someone, somewhere, will figure out that question).

Where I do see immediate value in these custom GPTs is for internal use. For example, over the past week, I set out to train a GPT to be a writer for my Toolscape website, which is a growing directory of AI tools. Building out a site like this is time-consuming, and AI has allowed me to 10x the speed at which I can review new tools and get content onto the site.

Anyone who has tried to use ChatGPT so far to create content quickly realizes that although initially impressive, the produced articles are lifeless and will be hard to rank with. It was with this in mind that I wanted to customize a GPT to help with article development, which could then be easily edited by a human. The main focus, though, was that I didn't want just generic content to fill pages - I wanted something that was in sync with my vision for the site.

I'd say we are about 70% of the way there so far. At this point, the introduction paragraph is typically weak. And, as with all of ChatGPT's features, no matter how much you berate or demand it, it insists on putting in a cheesy summary paragraph at the end. It may not for a few articles, but then, sure as hell, the "in summary" paragraph appears like a bad habit.

I've also been able to fine-tune the personality of the content, as I want the content to have personality. On this note, it took some time to cut out the hyperbole and sales pitch element of the article, and to this point, it has remained out of the produced content.

One final note on this GPT - as part of the training model, I am able to feed it in anything I have written in the past to teach it to mimic more of my first person conversational style of writing and I am hoping over time it will be able to create content indistinguishable from my own writing. ( I know that freaks many of you out but to me it's an ultra impressive feature )

A GPT Solar Power Calculator

One of the GPTs I have been most impressed with is the solar power advisor GPT I created. The nucleus of the project came from my desire to figure out what type of alternative energy system I would require to run my electric hot water tank, the well, and the pressure tank at my cottage. I'm not at the point where I wanted to go full-on solar - not to mention the significant cost involved. But running this expensive appliance, as well as its components on solar power, would not only save me money in the long term - but in the case of a problem with the power grid, would provide me the ability to have running hot and cold water regardless of the power being on through the grid.

Finding the answer to this project, though, was near impossible. There are lots of calculators out there - and even more people offering advice, but nothing really specific to my job at hand.

I have a 30-gallon tank that uses 3000 watts of power and requires a 240-volt power supply. The well and pump run off standard household power, while the pressure tank is hardwired into the wiring of the building.

The solution - create a GPT that was trained in solar power requirements, which honestly wasn't that hard. Once I had it operational, I gave it the following question (prompt):

"How many solar panels will I require to power my 3000 watt electric water heater that is a 240 volt powered tank?"

The following was its response.

"To calculate the number of solar panels needed to power a 3000-watt electric water heater, we need to consider a few factors:

Energy Consumption of the Water Heater: Your water heater consumes 3000 watts (3 kW).

Average Daily Usage: How long the water heater runs per day. Let’s assume it runs for 3 hours a day. So, the daily energy consumption would be 3 kW × 3 hours = 9 kWh/day.

Solar Panel Output: The output of a solar panel depends on its efficiency and the amount of sunlight it receives. On average, a standard solar panel produces around 250 to 400 watts of power. For this calculation, let’s assume a 300-watt panel.

Sunlight Hours: The average number of peak sunlight hours in your location. This can vary widely, but for this example, let’s assume an average of 5 peak sunlight hours per day.

Now, let’s do the calculations:

Total Daily Energy Requirement: 9 kWh/day.

Daily Energy Production Per Panel: 300 watts/panel × 5 hours/day = 1.5 kWh/day/panel.

To find out how many panels are needed:

Number of Panels: Total Daily Energy Requirement / Daily Energy Production Per Panel.

9 kWh/day / 1.5 kWh/day/panel = 6.

So, you would need approximately 6 panels of 300 watts each to power your 3000-watt water heater, assuming it runs for 3 hours a day and you have 5 peak sunlight hours. This is a simplified estimate, and actual requirements may vary based on factors like local climate, the efficiency of the solar panels, and the actual daily usage of your water heater. It’s always a good idea to consult with a solar energy professional for a more accurate assessment based on your specific circumstances."

As you can see, the response was pretty impressive, and when I ran that response past an electrician friend of mine, he said it was bang on for the requirements. In a subsequent response, it also explained exactly what I needed in batteries for storage and additional panels to handle the other electric requirements.

The point being that here are two examples of how I took custom GPTs and added value to my own internal needs and honestly, the sky is the limit when it comes to the applications that a person can use these for their own value - whether that be personal or even in their business.

Limitations and Failures

One of the areas where my GPT didn't do well after some initial training was in a project where I wanted it to evaluate page one search engine rankings and report back as to why certain sites had achieved their rank. For example, in a competitive niche like injury law for a certain city, there are many factors that contribute to top rankings, and most sites are highly optimized by professional agencies.

In less competitive niches - say for local embroidery shops in your hometown, the SERP results and the sites at the top often have far different logic for being there. In a study I did a while back, the clear factor for the top rank on the keyword was the age of the domain - whereas in a competitive local niche like HVAC, the age of the domain had little impact, and backlinks along with reviews had a much higher relevance. For injury lawyer keywords, content and local relevance had far higher weight.

The point being is that I wanted to see if I could train a GPT to scan the top three sites on a competitive search engine term and provide me a detailed analysis of why a particular site was, say, at spot two in search results, and what it would have to do to move to spot one.

The initial test was a complete failure, and I am faced with the decision: do I continue to train it until it works, or abandon the idea for now? From a commercial standpoint, it has definite value to past clients of mine - as well as any SEO agency or freelancer wanting to use or market the technology and design. So many ideas, so little time.

With that said, I'd definitely encourage you to upgrade your ChatGPT to the premium plan so you can start experimenting with your own creations. In the meantime, I will continue to bring you my projects and adventures with this fascinating new tool.

If you like my writing and honest approach to talking about AI and how it can be used in real-world settings, why not follow me here on LinkedIn so you don't miss any future articles.

• By the way, I spell checked and had GPT edit this article for grammar and punctuation - something I used to have to either do myself or ask someone else to do. Took about 15 seconds... The image for this post was also created by ChatGPT using Dall-e based on the following prompt " Create me a pixar style 3d image for my article discussing the custom GPTs I have created to this point. It should measure 1920x1080 pixels and be of high resolution for use on Linkedin.