AI Policy

This statement outlines my approach to using AI-based tools, primarily in the making of this website, but also in personal life and at work.

I am using the term "AI" here loosely to refer to various technologies—including but not limited to image, video, and text generation tools, chatbots, and vibe-coding tools—built with large language models at their core. Technology evolves rapidly, and my thoughts on this topic will likely change as I continue to explore its potential, limitations, and implications. I intend this policy to be a living document that will be updated as my understanding of AI evolves.

1. General principles

My approach to using AI-based tools is guided by the following principles:

1. Transparency: I will be transparent about when and how I use AI-based tools.
2. Accuracy: I will ensure that these tools do not perpetuate harmful biases or misinformation. I will critically evaluate the outputs of AI tools and make adjustments when necessary.
3. Continuous learning: I will stay informed about the latest developments in AI technology and best practices for its use, and I will continuously evaluate and update my approach as needed.

As with any technology, I understand that LLMs can be and are being used for nefarious purposes by bad actors. I also find the current bubble around AI concerning, with a lot of power concentrated in the hands of a few (young and male) "hotshots". I feel the potential of the technology to do good is being squandered, with the current focus on profit, power, and warmongering rather than ethical considerations.

The extensive infrastructure required to maintain the current trajectory of AI development also has concerning implications for the environment. I believe a delicate balance can be struck, but that would require Silicon Valley to grow a conscience, so I am not holding my breath.

With all this in mind, I have endeavoured to keep my AI usage to a minimum, but I would be lying if I said that I do not find its prospects to be exciting.

2. Use within this website

2.1. General usage

The current version of this website, which is built on Astro and deployed on Vercel (see /colophon for more information) has benefitted from AI-based tools in its creation. The raw HTML, CSS and the minimal JavaScript were written by hand. However, I have used the following AI tools to assist with checking the code for errors.

• The built-in Copilot integration in Visual Studio Code
• Google Gemini

2.2. Vibe-coding

In a previous version of this website, the following iframes embedded on my /now page were created/"vibe-coded" using Lovable: Trakt integration, Goodreads integration, and PlayStation integration. This is no longer the case; I have replaced these with simple text links to each silo.

Two tools that I use regularly in connection with this website were also made using Lovable. They are hosted at physicalmedia.thamara.co.uk and moria.thamara.co.uk.

2.3. Creative usage

I do not use generative AI tools for creating any content for this website. All writing here is 100% human-generated. Images have been sourced from royalty-free stock photo sites, or used with the permission of the respective owners.

3. Usage elsewhere

Outside of this website, I use AI-based tools primarily for research and learning purposes. I am cautious about the potential risks and ethical considerations associated with AI, and I strive to use these tools responsibly and minimally. I have discovered that tools like Gemini are particularly helpful in distilling complex information into more digestible formats. I use the "see thinking steps" feature to better understand how these tools arrive at their conclusions. This was a helpful tip shared by Marisabel (the creator of the excellent Konfetti Explorations website) recently on a podcast.

At work, due to the nature of the output demanded of me, I have to separate my personal principles from my professional use of AI tools. I have used generative tools for drafting emails, social media post captions, and images.

Versioning

I use the following method of semantic versioning for this policy: Major.Minor.Stylistic