As I look back on the first half of this year, it sometimes feels like being on a rollercoaster. Why is that? Keeping current with everything AI, while at the same time becoming a practitioner of the art, certainly keeps me more than occupied (even more so when doing this as a side activity).
As a break in my current series of posts on AI's history, I wanted to share with you some of my current thoughts of where I see the field of AI going, and some of the practical applications where I use AI.
When I initially got into the topic of AI (see my first post) I was still unsure and not convinced about the real value of AI. Like many others I felt this could still be an impressive technology but fundamentally constrained (remember the words I used ... a polite parrot with a PhD). Constrained because it could create and generate new content only by rearranging and resampling already available text and content. Therefore it was an open question if there was any "understanding" involved, and the majority of opinions thought not. I now believe this becomes more difficult to maintain. I am not claiming that AGI (Artificial General Intelligence) or Superintelligence is right around the corner, or even already here. But the capabilities of the latest AI models are genuinely impressive. And the evolution and growth does not seem to slow down. A few recent developments were important for me. AI proving (or disproving) long-standing problems in mathematics. AI models finding software vulnerabilities and exploits, often in software which has been around for years or decades. And the most recent news with some of the more advanced models escaping out of their sandbox environments (granted, there may be an element of hype or marketing involved here as well. We do not yet know) seems to point to an ever-increasing level of sophistication and reasoning displayed by these models. No, these are not stochastic parrots anymore.
But what really convinced me was my own experience using AI (in my case Claude, but it could be any other application of your own choosing). Once you start using these tools in a more structured setting (not just the occasional chat) the possibilities are limitless. I am not using Claude to manage my schedule or my e-mails (although that is still an option for the near future), but I use it as a very knowledgeable, and ultra-patient, assistant and helper to accomplish tasks that I would not have been able to complete on my own. Following are some practical real-life examples where I used Claude to achieve a task.
The first example is designing and building my website (where you are reading this post): I know a bit of html and css (both languages needed to design webpages) but lack the detailed and in-depth knowledge. So Claude helped me to implement specific features and styling elements, to ensure a consistent look and feel, and gave some helpful suggestions in writing the content. The writing is still mine, but Claude is a great editor, and I see myself as the final editor-in-chief. It allows me to focus more on what I want to say and express, without getting bogged down in the many subtle problems and issues. Claude provides me with the fixes, and is an excellent tutor and explainer of things ... and it never gets grumpy or loses its patience!
A website has to live somewhere, and I chose to get my own VPS (Virtual Private Server), basically a computer I rent and which is already connected to the internet. It's infinitely scalable (you can always rent more memory, disk space, CPUs, GPUs) and some of the basic security and maintenance functions are already provided as part of the service. But everything else is your own responsibility. I know some basic Linux (the operating system typically used for these servers) but again lack the detailed knowledge. Claude provided me with a step-by-step plan, sat next to me as I set up the server, told me what packages and tools to use to run the webserver (I use Caddy in case you're interested) and patiently explained to me how it all worked, what commands to use, the meaning of these commands and other useful context. It occasionally got things wrong. Sometimes it did not work. But here also Claude works with you. You paste the error messages and logs from the server back into Claude, and it almost always does a great job in diagnosing the problem and providing a solution. This is where I started to feel the power and the capabilities of the technology ... it's more than re-arranging words and tokens, there is a reasoning capability which is genuinely useful. The result is that I am now the proud owner of my own website, hosted on my own rented machine, and in the process I learned a lot about servers and websites.
Finally I wanted to get my hands dirty and build my first AI app. Inspired by Ed Donner (see my Resources section), I decided to build my very own AI Twin, a chatbot which knows all about my professional career and skills, and can answer questions about (professional) "me". Building the app was quite straightforward and I knew how to do this (on my own local computer, a bit of Python coding and using a Jupyter notebook). But I wanted this app to be publicly available by clicking a link on my website ... and I was off for another learning adventure with Claude by my side. Installing an app on a server was brand new to me, and getting all the details right (managing user authorisations in Linux turns out to be an arcane art and practice) took me most of a day (and quite a bit of coffee). I was sceptical I would ever make it when we went down another rabbit hole with Claude, but it kept track of where we were, kept explaining and analysing things, and to my surprise in the end it actually worked.
So what has the above taught me? The ideas and the vision of what to build or achieve are not coming from AI, they come from Wim's wet biological brain. But implementing my vision was impossible without AI. The vast amount of knowledge and information the tool has access to, together with the ability to think through a problem and propose a solution, has impressed me greatly. Of course, I know that coding is probably the single largest area where AI tools have already made a significant impact, and I can only confirm that. It does not mean that AI will be as successful in other areas. Time will tell. But I also know I have barely scratched the surface, and I am still very conservative (e.g. I usually do not ask Claude to make edits directly on my behalf, as I want to ensure I learn while doing). But it might well be that in another few months I have got over my reluctance, and just ask it to make changes directly in my code. I will definitely keep you posted.
Where the journey takes me next is something we'll discover together. There is still a huge potential out there. The tools keep evolving, and I truly enjoy experimenting with them, in the process learning a few new skills. I hope to also educate and inspire you to take a crack at it, or at least awaken your curiosity. Write to me with your own stories, and I would love to hear from you what else you want to explore together here.
My next post continues the story of the development of AI, right where we left it off in the previous post. Don't forget to subscribe to updates if you want to follow along here.