My experience with ChatGPT
I tried it out, here there are my thoughts.
Friday, January 20, 2023
During the hype for ChatGPT, I had a moment where I played with it A LOT.
I experimented with a lot of things, from programming challenges to ethical questions. Then I wanted to try one of the main features of ChatGTP: content creation.
There are 3 blog posts here on the blog, that were generated by ChatGPT:
- Optimizing Typescript for Enterprise Projects with tsconfig and eslint
- Real World Use Cases for Firestore
- Mastodon - A Decentralized Social Networking Alternative
I think that is very clear that those posts were generated by an AI.
Don’t get me wrong, the work of OpenAI and the skills of ChatGPT are amazing, probably the best performance that can be produced today.
I mean, how they are written, the structure… I don’t know exactly what it is, but everything is somehow impersonal. There is something cold and clean in that content, that kind of cold and clean typical of machines.
I don’t know, maybe I saw too many cyberpunk and sci-fi movies and I am influenced, or I am not objective because I know that those posts are generated and I spent some time fine-tuning the generated posts because I had to understand how I had to ask to get what I wanted (btw, the content comes 100% from ChatGPT, I just removed some cringe jokes. More about it later in the post).
By the way, this is another interesting point: how can you get the best results from ChatGPT? How you ask is a fundamental part (yes, I found out that we humans still have a main role, also in AI-generated content).
How to ask?
The way you ask something to ChatGPT is the key to getting the content you want. It is not easy, though, to get the right question, and is a try-and-error process.
I started with a simple question, like
Generate a blog post about Firebase.
And the answer was too generic, short, and not like I wanted to have a blog post on my blog.
Then I started to fine-tune my question, trying to be more specific.
The question I came up with at the end, looks like this:
Generate a blog post about Firebase. Consider the following:
- The blog post is written for expert software developers, do not explain simple things
- The blog post is intended to explain technical things, not to sell something
- Include some code examples that people can copy and paste
- The blog post conclusions must include a funny phrase
- before the content, put a tl;dr to summarize the post. Maximum 50 words.
Why those specific indications?
My question explained
The blog post is written for expert software developers, do not explain simple things
The first generated blog post was explaining really simple things in really simple words. It is a good thing, but for my purpose, it was way too simple. It didn’t go deep enough into the topic.
Include some code examples that people can copy and paste
I just wanted to put something more appealing for the developers reading the post. Moreover, forcing ChatGPT to put some code, it had also to go more technical in the content. And that is what I wanted.
The blog post conclusions must include a funny phrase
The tone of voice of the first posts was quite cold. I wanted to put some “humanity” by inserting a joke at the end. It did not always work, sometimes the jokes were quite awkward 😅 I gave ChatGPT the chance to generate something better, but it looks like the humor of ChatGPT is a bit cringe. I removed it from the final result, only Optimizing Typescript for Enterprise Projects has an example of those jokes. And this was the best one 😅
before the content, put a tl;dr to summarize the post. Maximum 50 words.
Like the joke, it didn’t work always very well. It was often a summary of my question, rather than the summary of the post. In the end, I had to remove it from all the 3 posts. I leave it here though, in case you want to try it out.
The result is the blog posts that you can read in the blog.
I leave my “Question” here, in case you want to try it out.
Reaction from a non-tech person discovering ChatGPT
A friend of mine is writing his Ph.D. and talking about his work, I said that he could use the capabilities of ChatGPT to make progress on the research he is doing. At first, he did not understand what ChatGPT could do, and he thought to be in front of a more fancy Alexa or Google Assistant.
Then, I challenged him to ask ChatGPT to write a piece of text about a very specific topic of his research. Something that for sure, was hard to generate without very specific and deep knowledge of the topic.
Of course, ChatGPT generated a quite good text, and he was shocked.
We were at the table together with our wives. His face went from skeptical to amazed in some seconds. He said:
“…can we please go in the living room? We must talk.”
He could not believe his eyes.
ChatGPT is quite new and people outside the tech industry still have to discover it and understand what it is and what is capable of. But this little personal story tells me that the implications and the use cases for “normal” people (people who are normally not interested in tech things and artificial intelligence) are incredible and devastating. If the devastation will be positive or negative, depends on how we will use this incredible technology.
It was interesting to see how the human contribution to the ChatGPT content generation, the question you ask to get the content you want, covers a fundamental role in the process.
It looks like ChatGPT can do anything without humans, but, without a human that asks the right answers, the content generated by ChatGPT is quite disappointing.
As I said, the work of ChatGPT is amazing, and with the right questions, the potential of this tool is astonishing.
Just remember, that the questions are important 🙂