I'm not impressed with the new GPT's that everyone is talking about.
Here's why...right now they only really work when you upload a load of data to feed them. But getting them to perform simple tasks not functioning as they should, yet.
They were announced last Monday, came to most people's screens around Wednesday and there was the usual race on to show off GPT's people have created.
Casting aside my inner competitiveness, I decided to wait and ponder over what I would like my first GPT's to do.
I grasped the concept from Sam Altman's speech: they are specialist GPT's designed to act as an assistant for a single use case. Your custom GPT to focus on one task, that you don't have to build out everytime by prompts.
I came up with 2 good ideas over the weekend:
- Build a GPT to find me guest podcast opportunities (by the way, I'm available and I'd love to be on your pod if you think I'd be a good fit!)
- Build a GPT to monitor the price of home heating oil and inform me when it goes below a certain price point
Both of these I need.
How did I get on?
1. Build a GPT to find me guest podcast opportunities
Here it is.
Too slow. Takes ages to browse with Bing. My preference for accessing real time information is to go directly to Bing as I find the chatgpt browse with bing feature really chuggy. After a lot of waiting, it found 1 podcast. And it's not that relevant either.
I tried to get more specific and give it examples of podcasts I have been on. I asked it to scan the other guests and see if they have podcasts that might be of interest. It failed at that.
2. Build a GPT to monitor the price of home heating oil and inform me when it goes below a certain price point
This should be easy. Look up oilprices.ie and tell me when the price per 500 litres goes below a certain price point. I even inputted the price point into the instructions and it still didn't work. By the way, the price today is the lowest it has been since mid August - I found that out myself, using my eyes, after visiting the website in despair with ChatGPT. Here's the oil price GPT.
I turned to YouTube and watched several videos of creators who were all dying to show us how it's done.
I noticed that most of them spend about 12 minutes marvelling at how ChatGPT helps you pick a name for your GPT, and it even comes up with a logo etc. Big swing.
The only ones I've seen that work are GPTs that are trained to be marketing assistants.
To me, they just look like ways of summarising all your persona and content creation prompts into one place, and then getting your blog posts, etc made from there.
I don't see the value. But I tried again, this time uploading data on all the course feedback received this year.
3. Build a GPT to review the course feedback
I configured it to analyse data uploaded in csv format. This is what it produced. Slightly better in that it actually does what I want it to do, but it's still just giving me the exact same results as I would get if I analysed the data via code interpretor on ChatGPT.
In conclusion, I think GPT's have potential but for now they are just containers around what you may already be doing with your prompts.
I'd like to remind you about the demo we all marvelled at back in March when GPT4 was being launched. The one about how we can draw a website on a piece of paper and it will be turned into code!!! Yes, I was excited about that too. We've only just got GPT Visual where it can read what's written down. So I conclude that OpenAI gave us a big shiny demo last week of what's going to be great in the future. They generated so much coverage and excitement and they are really good at doing that.