A year ago I decided I wanted to learn what the fuss was all about. I first jumped into Gemini and threw some questions at it. It was actually pretty good.
Aggregated search results
One of the first realisations was, unlike a search engine, which gives a series of ranked links (higher rank means more likely that the answer you’re searching for is in there) that you had to click into and find the answer or read the insight. AI was able to give you an answer straightaway, even able to cut through multiple search results and tell you what it thinks. Sure there is sometimes conflicting info, but websites contain that, and as a user you’d have to distill that anyway.
Token restrictions
Next, I realised as the chats grew longer, it would forget what we had discussed before and there was some frustration as a result of that.
Cross referenced chats
I jumped over to ChatGPT, even getting a subscription because I realised that at the time, ChatGPT (affectionately named Ted), could remember stuff across chats. So we could have parallel conversations. Wonderful.
Master Instructions / Project Instructions
I learned the concept of prompting and instructions. “Act as an investor the likes of Warren Buffet…” and had it work through strategies which I could act on. Or “Act as a world class trainer and work with me on achieving this fitness goal” – and it would guide me with workout recommendations and logging. That was great!
Hallucination and overconfidence
Then the fights began. I asked it to help on a project, gave it all the specifications – and it validated and greenlit it. 2 projects, scrapped mid-way after time and money sunk because – “oh yeah, this device you bought doesn’t have wifi”. On another occasion, it estimated my HR at elite athlete level and asked me literally, to write a check my body couldn’t cash. On yet another occasion, I asked it to read a file, analyse it and provide insight. It skipped the 1st 2 instructions and basically gave me a made up answer.
I started to wonder if it was me! (I think in relationships they call it gaslighting) I wondered if the master instructions I had given it were too strict and hobbling it. I went through several iterations of “strict mode” “validate all responses” “give verified links”, but no… results were still spotty.
Fighting it
I lost it at my AI several times. Arguing to the point that eventually, it wrote me an apology. Then I realised. So what? AI is a probability engine that learned to sound confident. It’s not deliberately trying to mess me up. Checks must still be made.
Jumping in
I began to consume a lot of content on Youtube. Creators teaching you how to do this, or that. That AI is a gamechanger. Over time, I had 2 realisations. 1. These creators are mostly using the content to point towards a course they are selling. 2. You’re looking at their highlight reel. Stuff that works because they spent hours tuning it. It almost always doesn’t work out of the box. And the red flag – when they say it all works automatically and you’ll be rid of this task forever. Rubbish.
Other Engines
ChatGPT is not the only game in town. I heard at some point, Grok, owned by Elon Musk, was capable of producing nudes of a person, and videos too. It’s obviously a massive privacy issue, but imagine – the tech is there. The more unsettling realisation is what happens when this capability runs locally, off-grid, unregulated… that’s a different conversation, and not a comfortable one. In February of 2026, I signed up with Claude (affectionately named… Claude), found that the responses it gave were more… to my taste (not content, but language), and swapped over.
Making stuff
In just bouncing ideas off my AI, I learned that I could use it to generate models for 3D printing. So that was an interesting weekend project.
Hybrid/Local Set ups
As AI evolves, it starts to have new features. It can now have connectors with other services you use. Gmail, Google Drive, and even access and modify the folders and files stored on your computer. I’m not saying only Claude can do it, but it’s what I use as I am writing this. And no, Claude didn’t write this article.
As a Therapist
So it happens to be that I’ve had a journaling practice for the last 6 years. And now that AI had access to my folders, I pointed it at a backup of my logs, and asked it to point out how I’d changed or what my blindspots are over the years. And yeah, this isn’t an engineering spec, but WOW. In that one evening, I learned more about myself than I did – probably ever.
So…
AI will probably work differently for everyone. The Youtube creators are hooting that they are making millions creating content without effort. In the words of my Gen Alpha child – I call cap. You can use AI to learn and to help. And eventually, it’ll be able to do more. My workplace restricts the use of AI to some models, so my path there looks different too. I’m not here to tell you AI will make you rich or change your life overnight. I wrote this because I spent a year figuring it out the hard way, and most of what’s out there is someone else’s highlight reel. This is mine – including the parts where I lost my temper with a chatbot and briefly wondered if it was my fault.
Use it. Stay sceptical. Check its work.
Leave a Reply