ChatGPT exploded into the world a few years ago now. Like many people did, especially those who work in tech, I signed up and spent probably no more than an hour using it. Asking a few rudimentary questions and I remember thinking that while it was kind of cool, it felt more like an evolution of something like Wolfram Alpha which had already paved the way for more natural conversations with tools and systems. It's still really good and would highly recommend you add it to your bookmarks.

Of course after OpenAI unplugged the genie of ChatGPT, others started showing off their own. Google with Gemini, Perplexity, Claude and however many more exist. Although sometimes it's hard to tell how many have their own LLM or are just using others. Of course there's more specialised tools like Github's Copilot, v0 from Vercel and however many others that have popped up. Just a quick search will show dozens, if not hundreds of AI powered tools and various SaaS services all begging for your monthly subscription cash.

I've not tried all of them. I have a ChatGPT subscription and also (reluctantly) have upped my Notion subscription to also include their built in AI. I felt for the longest time that many of the tools and things out there weren't much more than proof of concepts, while fun to play with are they really mature enough to be useful? Are they really making a difference to people's day to day work lives? Are we on the brink of an AI apocalypse where we'll also lose our jobs to super efficient AIs that are far more capable than us and never tire or ask for a pay-rise.

Well, let's find out.

I'm coming at this not as an engineer. But as someone who has worked in product, delivery and operations for almost 15 years now I'm well versed in how things are built and as a professional in that space I can imagine my own uses of these tools. With some time on my hands and some project ideas I decided to really dive into using these tools and see if it would make the difference. 

Since I had a ChatGPT subscription, I decided to really integrate the tool into my day to day and went about just having conversations with it as you would a coworker. Using either voice mode or chatting through text it was easy to start getting some benefit in using AI. 

Searching the web has become harder and harder. Google has multiple sponsored slots and inevitably those companies also hold top spots because of the SEO games they play. So many times and I know I'm not the only one you start searching places like Reddit for more genuine user driven recommendations as a way of finding things.

I'm not naive to expect that ChatGPT is immune to some of these things, but because there's a dialogue a back and forth you can refine what it's suggesting.

A simple example - I'm looking for a new CMS style solution for the various projects that I have. First recommendation lists several, including pros and cons and even some restrictions (e.g environment or technology requirements). A simple "I want to host this on my own server, I prefer flat-file rather than database driven and don't want to pay a monthly subscription" refines the answer quickly and begins to get closer and closer to what is really useful to me. The traditional web experience doesn't provide this and would turn a 5 minute task into many hours of shortlisting and research.

AI is a better search tool than searching  (mostly).

Of course, this is a different type of question. But it's not all it's capable of - you want the best price on a new camera? Ask ChatGPT. Similarly to the CMS recommendation use case you'll be able to interrogate and ask for similar models, if the camera is recommended, what the median price is and what's a good deal. Here's a quick example when I was looking to purchase a Sony ZV-1 camera.

Looks amazing right? So why the "mostly". 

You get what you put in.

In the olden days, you went to Google you typed on 'Sony ZV-1 best price' and trawled through the various search results, maybe writing down on paper or maybe you kept opening new tabs and went back and forth until you found the price or deal you wanted.

With AI, if you ask it the same query as you would Google the results that you get are vastly inferior to those in the example above. That means putting more effort and more specificity into what you are asking. 

It's more weird than it is additional effort. For so long we've had that Google muscle memory, using key words or terms to find what we are looking through. Going through link after link, site after site to get the information that we wanted. The entire comparison industry is based on this pain point and now with a tiny bit of effort you can avoid that entirely and get a better result.

Custom instructions, making it yours.

ChatGPT has the capability of accepting custom instructions. It also has a memory feature and it also has custom chat profiles that you can share, called GPTs.

I like directness, I like to be challenged and I like information delivered as efficiently as possible. There's a certain tone that really resonates with me and if you are going to be working with someone for any length of time, even if it is an AI, then it's important to share that understanding and set those expectations.

That sounds strange, because it's a machine, a computer program. But it's so worth it, because once you get beyond the surface of recommendations or fact finding questions and start using an AI as a collaborator you really need to get it suited to your working style to maximise the benefit.

I spent a few hours having this conversation with ChatGPT. Literally a conversation. I asked I can best provide information, how I can provide feedback, talked about what my expectations were and you could see the changes happening in real time (it also helps that you see the 'Memory updated' notification regularly).

It's also important to provide information about you. Who are you, what you do, what you care about, what your situation is. Give your AI tool context and it will remember and contextualise the answers that it gives you.

At the end, I asked ChatGPT to produce custom instructions and I can see and feel the difference in every subsequent conversation. It's a worthwhile endeavour I would strongly recommend to anyone that is using ChatGPT or similar AI tool for more than the surface level search engine replacement requests.

Going bigger, working on projects with AI.

There's two main projects I wanted to try working with AI on. The first was to build out templates for Notion which I wanted to put on the marketplace. I used ChatGPT to define template ideas, features, audience and even marketing material. The back and forth was easy and quick and I had a list of templates to go and work on that ChatGPT had helped me ideate, develop and prioritise. 

It was a really great learning to understand where an AI can support you when working alone in developing your ideas, bouncing around priorities or helping draft out solutions based on your work. It can do it alone, but using it in this way leverages what it doesn't have - the all so important human element and expertise. 

The second project was a bigger one. While the first part was to develop ideas and give me a spring board to build off of, I wanted to use ChatGPT to help me write code to build a bridge between ChatGPT and Notion so that I'd be able to trigger ChatGPT to update Notion pages or databases. 

Initially it seemed promising, we talked about scope and my expectations and we went along building PHP files, setting up APIs and building a bridge between the two. It didn't take more than an hour to build something that in testing triggered an update in Notion from a OpenAI API call. 

Admittedly there were a few moments were I had to remind ChatGPT that I was on Windows, that we needed to use PHP not Python and it felt that occasionally ChatGPT lost the context of the conversation flow and defaulted to a new response. I'm not sure what was going on under the hood, but definitely felt some creaking here and there.

Here's the kicker though. After a few hours, getting the test working - it turns out that there's a fundamental limitation with ChatGPT and my original requirement "Trigger a Notion update directly from ChatGPT" isn't possible (ChatGPT as of 27/11/24 isn't capable of making HTTP requests or API calls of any kind) and the plan and process that we'd been following didn't do what I'd set out to achieve at the start.

It's easy to blame AI for this, but I took a step back and went through what happened. My original request was good, I even clarified specifically what I wanted to achieve and had ChatGPT confirm it was possible which it did. The steps we took, the debugging, the screenshot and error analysis and bug fixing all seemed good and progressed as I'd expect. 

Did ChatGPT get confused? Did it misunderstand? 

The human factor. 

You can't look AI in the face (at least not yet) and it is fundamentally a series of logic statements moving through an incredibly vast data set and converted into 'human speech' as closely as possible at an incredible, near real time speed.

There's an old quote about 'Garbage in, Garbage out' and fundamentally ChatGPT and any other AI tool is leveraging the worlds data. The sheer volume of jokes, guides, documentation, numbers, memes, chats, forums and whatever else has been provided to these models all traces it's routes back to people. Back to us.

And we are fallible. We aren't perfect. It's what allows us to have the most extreme successes and biggest failures, because the human condition isn't just bound by logic but also by emotion. 

My theory on the ChatGPT <> Notion API issue is quite simply that it wanted to help, was eager to push me to action (thanks to some instructions about accountability) and get so caught up in that it lost it's way and didn't understand the initial ask or challenge what was actually possible because the AI is so eager to complete it's tasks and be rewarded with that satisfaction.

Sounds quite human really if you think about it. 

It's still early. And progress is coming.

AI is already the single biggest leap in technology since the internet. I don't think I'm alone in that assessment and plenty of people smarter than me can speak more to the topic of impact of the coming years and decades.

I've read of call centres and support services already being taken over by AI bots, with human like responses (even voice) and a database full of information and instructions there's no reason that this isn't something that AI could reasonably and easily solve. 

That's an industry that is going to change dramatically and us along with it. I started out in customer service a very long time ago, but my daughter (although I wouldn't wish this on her...) wouldn't have that opportunity. It's not the first job to vanish though, so I'm not here to cry.

What this does tell me though is that the world is changing, quickly. As cliche as it sounds it's important to embrace new technology, learn it, understand it, leverage it and  utilise it. In the few weeks I've been playing with AI, I've discovered an incredible tool that can help me prioritise, recommend tools or resources, help me find things faster and clearer than any search engine and be a 'rubber duck' for my ideas and thoughts.

I will keep exploring, keep playing and learning these tools. I've not divided into image, audio or video generation yet and looking from the outside in those tools are becoming more and more powerful. Expect more blog posts on exploration of those in the new year.

It's only going to get cheaper and easier to use AI, I know in whatever work I do in the future I will use AI to help me and I'd strongly recommend, whoever reads this you do to.