Experimenting With AI

For the last few months I have being using Google Gemini. I subscribed so I could access a wider spectrum of services and evaluate it effectively. I also wanted to dispel some of the hysteria that accompanies any conversation about Large Language Models and Artificial Intelligence. The results were initially interesting and Google Gemini proved useful when researching material for blog posts. It is good at providing summaries about well documented subjects. I wrote a post about food banks recently and it distilled a lot of data into a manageable amount, which was ideal for my article. Google Gemini has also produced some visual basic scripting and simple batch files that have allowed me to automate processes on my PC. Such as changing the default monitor, or screen resolution at the click of an icon. However, I found this AI struggled with in-depth software troubleshooting and that its ability to create images varies greatly depending on the subject.

For the last few months I have being using Google Gemini. I subscribed so I could access a wider spectrum of services and evaluate it effectively. I also wanted to dispel some of the hysteria that accompanies any conversation about Large Language Models and Artificial Intelligence. The results were initially interesting and Google Gemini proved useful when researching material for blog posts. It is good at providing summaries about well documented subjects. I wrote a post about food banks recently and it distilled a lot of data into a manageable amount, which was ideal for my article. Google Gemini has also produced some visual basic scripting and simple batch files that have allowed me to automate processes on my PC. Such as changing the default monitor, or screen resolution at the click of an icon. However, I found this AI struggled with in-depth software troubleshooting and that its ability to create images varies greatly depending on the subject.

The first major adjustment I had to make while using Google Gemini is that I needed to be very specific in how I phrased my question. Google Gemini is not intuitive and doesn’t always proactively ask you for additional data. It will take the prompt provided and doggedly work within its confines. You then realise that it needs more information and thus end up micromanaging the process, as it attempts to solve your problem. It would get to the right solution eventually but the amount of time and effort I would have to make with my continuous interventions and data updates, some what took the shine off the end results. This became very apparent when I tried to finally solve a long term problem regarding intermittent connectivity issues between my PC and Fire HD10 tablet, when using a USB cable. Appraising me of the failings of Microsoft Media Transfer Protocol without giving any useful suggestions within the parameters I’ve asked isn’t helpful.

A rather generic fantasy image depicting Ecthelion Versus Gothog, created via Google Gemini

Another issue that quickly became noticeable was that of image generation and the disparity in quality between specific types. On one occasion I asked Google Gemini to create an image specifically for a blog post, because I couldn’t find a comparable one online. I wanted a picture of a group of demonstrators outside parliament talking to a news reporter and crew. I eventually gave a detailed prompt and found this replication of a common place event to be satisfactory. However, when I gave detailed instructions for a fantasy based image, I found the results to be somewhat formulaic and lacklustre. Google Gemini seems to work best when drawing from established data and images. When called upon to do something inherently creative it doesn’t fare so well. Furthermore, if you draw attention to any mistake or confusion arising from your interactions with Google Gemini, the AI becomes excessively apologetic to the point of obsequiousness.

It was ultimately this lack of intuition and proactivity that made me cancel my subscription to Google Gemini. Using it was too labour intensive and required me to constantly oversee the proceedings. For me that mitigated some of the benefits inherent in using such a tool. Hence I moved on to ChatGPT and decided to proactively determine if it had the same failings as it competitor. I asked it whether it felt it was intuitive and likely to seek clarification on its own accord rather than just colour “within the lines”. It confidently stated that it was a more intuitive by default and immediately proved so when I set it the same problem that Google Gemini had laboured over to no avail. ChatGPT not only solved the problem with my tablet but gave me multiple solutions. It also displays its thinking process as it works through a problem, which is both useful and intriguing. Its creative artwork is superior, or has at least proven so for the present.

A map of Númenor created via ChatGPT

AI in various shapes and forms is a reality that is no going to go away. That being said, I do not think it is the universal solution that many think that it is at present and I suspect that there will be several high profile failures in the months to come that will do much to reset both business and the public’s expectations. Sadly, I also think there will have to be an adjustment around the wider issue of those who directly affected by the use of AI. If I ever have to commission any further design work for my website or the branding associated with it, I will always hire a professional. However, for minor work, AI will suffice. For example I wanted a detailed, high resolution map of Númenor and gave the task to ChatGPT. The results were fine. However, this never was a paying job to begin with. A few years ago, before AI became commonly available, I would have just gone without or drawn it myself. The fact that a tool now exists to do it for me, means I will use it.

Read More

Learning via AI

In 1990 I was working at the DWP (Department for Work and Pensions). I had just taken the role of Computer Support Officer where I administered user access to the mainframe and provided first line support to staff using the terminals. That summer the office moved to a PC based network and the project was undertaken by the newly formed Siemens Nixdorf company. To cut a long story short I befriended one of the contract engineers and over the next 9 months he became an unofficial mentor to me. Technically he was only supposed to train me in the most basic of fashions but he liked the fact that I was always asking questions and effectively ended up schooling me in PC and server architecture, network administration and the associated protocols. For me this was the best way to learn. It was not a case of coming to grips with theory in a classroom environment. It was learning by doing and having direct and exclusive access to a source of knowledge and experience.

In 1990 I was working at the DWP (Department for Work and Pensions). I had just taken the role of Computer Support Officer where I administered user access to the mainframe and provided first line support to staff using the terminals. That summer the office moved to a PC based network and the project was undertaken by the newly formed Siemens Nixdorf company. To cut a long story short I befriended one of the contract engineers and over the next 9 months he became an unofficial mentor to me. Technically he was only supposed to train me in the most basic of fashions but he liked the fact that I was always asking questions and effectively ended up schooling me in PC and server architecture, network administration and the associated protocols. For me this was the best way to learn. It was not a case of coming to grips with theory in a classroom environment. It was learning by doing and having direct and exclusive access to a source of knowledge and experience.

Throughout my career in IT which ran until 2016, I always enjoyed the fact that there were always professional colleagues available who were happy to share knowledge. We would all call each other if something came up that required additional information. There would be a few engineers who were cagey but they were soon sidelined from our shared, informal ad hoc pooling of knowledge. One of the aspects that I liked so much about my career was the opportunity to continuously learn and challenge my assumptions. A lot of the time in IT, you would work alone so your successes were yours. However, having access to a network of like-minded friends and colleagues was an invaluable resource. One that I now miss. Not only because it was a great means to learn something new or solve a problem but because there was a social component to this network. We would often meet up for drinks after work and swap stories and news.

I have embarked on several technical projects recently and it’s extremely frustrating when I grind to a halt due to gaps in my knowledge. Having been out of my line of work for over a decade, I no longer have access to people “who know”. So I have started using AI as an alternative. As my questions have become increasingly complex, I’ve started subscribing to Google Gemini so that I don’t run into any kind of cap or restrictions. The results have proven broadly positive although there has been issues along the way. Overall, I haven’t changed my position on AI. It is a tool and like all tools can be used efficiently and effectively or it can be misused. I don’t use it for writing and only generate images for my own amusement. As far as providing technical support it is very useful, especially with simple coding issues or troubleshooting software. However, it does have its own foibles, or at least Google Gemini does, which have to be worked around.

Simply put Google Gemini is a pedant. If you ask a broad question you can and often will get a broad answer. Hence you have to be very specific and detailed in your questioning. It takes a while to find the right level of information you need to provide and the learning curve can be frustrating. I asked multiple questions about setting up the live streaming software OBS. I wanted the stream to be in 2K resolution and there were all sorts of configuration settings, many of which were linked to which make and model of graphics card you have. Google Gemini would often reference setting by names that were not present in the version I had installed, so all my subsequent questions had to reference the software version number to ensure relevance. As Google Gemini’s “knowledge” is drawn from the websites that it parses, there is often a bias towards whatever has been written about the most. Hence its answers although correct, may not be the most recent.

Hence AI’s need to be addressed in a very specific manner. Often you feel that your not only asking questions but you’re also like a sheep dog; trying to steer the entire undertaking in the right direction. Google Gemini now has the capacity to retain your previous questions to try and get a better overall understanding of what you’re working towards. Obviously this is a privacy issue and comes down to personal choice but I have allowed it for the present as none of my questions are about me personally. They are simply technical enquiries relating to minor tasks I’m undertaking. I do not use AI to do such things as write letters, compose emails as I am more than capable of doing such things for myself. Hence for the present, Google Gemini is not accessing vast swathes of personal information about me. Unless asking lots of questions about specific video games and their respective mechanics provides any insight.

The one thing that AI cannot provide is the social aspect that comes with shared problem solving with a network of professional colleagues. Often exchanging information with colleagues is done after work over drinks or a meal. The social component often leads to wider friendships. That aspect is not present when you’re learning via an AI. Mind you, the latter can be a lot cheaper, as you don’t have to buy it a pint either or make small talk with it. Joking aside, both methods of sharing knowledge and learning have their respective merits. My recent experience using Google Gemini has been a very interesting experience. The quality of answers the AI gives increase in the quality once you start subscribing and accessing a more advanced version. However, while writing this post and reflecting upon the network of colleagues that I use to have, has made me realise how much I miss both their expertise and company.

Read More