AI Manifesto

Generative AI is a very cleaving subject where we tend to put people in one basket or the other.

You might find my position sharply against the use of generative AI in many contexts. It is true. As of today, I am highly concerned about its negative side-effects and ethical concerns. I want to avoid the use of generative AI whenever I can.

Yet, I refuse the label "anti-AI" or "luddite".

While I am far from being an expert, I have done and continue to do my homework in using the technology. I am approaching generative AI with skeptical ¹ a stance, but with what I believe is a balanced and fair mindset

The following contains my current views and thoughts on generative AI based on my own experience with the technology.

I remain open to discussion and my opinion is always subject to change.

In my opinion, generative AI has currently far more negative side effects than positive effects.

Therefore, I approach the use of generative AI with a "less but better" mindset. I will restrain from using it unless there is a clear and undeniable reason to do so.

I am (very) concerned about the current landscape of the AI 'labs'. I do not support any of the centralisation of the technology campaign conducted the current frontier 'labs'. I am a proponent of open source models, especially smaller, on-device models.

I am very concerned about the social, societal and environmental impact of the current usage generative AI.

Beyond that, I also believe that the use of generative AI will erode my critical thinking and cognitive skills. Therefore, I believe that the unrestrained use of generative AI is counter productive over the long term for the individual.

All the writing on this website is mine. None of it comes from generative AI.

I have used LLMs to review blog posts in the past, but I no longer do so. I now prefer to embrace my weaknesses and work on them.

I do not use generative AI for any writing communication, even at work. Writing is thinking; I write to lay out, organise, express, challenge and sculpt my thoughts.

If it's not worth the writing time, it's undoubtedly not worth the reading time.

I use generative AI for assisted coding for my current role where it feels economically advantageous for my employer. This isn't always the case: I try to balance short-term and long-term productivity.

While coding agentic tools can undeniably speed up part of the software creation process, I believe these are often short term gains. There are numerous issues arising in the use of agentic coding, especially in the medium to long term. For example, the unrestrained use of agentic coding in work environment trades short term speed for cognitive debt. I see them as I see sprinting; you don't win a marathon by sprinting - yet sometimes, you got to sprint.

When I use agentic coding tools, I use it as a 'dumb' faster coder.

Not as an engineer, but as a coder. A very efficient one.

It's purely about speed of producing code. The designing and engineering part is conducted upstream without the use of generative AI. Sometimes at dawn while sitting on my surfboard waiting for a wave.

My current employer supports but does not enforce the use of generative AI; it's purely my own set of rules.

I believe there is a moral obligation in using a tool if it benefits my employer. It is my responsibility to terminate my role if I think that the values of my employer are diverging too far from my own values framework. In other words, while I can voice my opinion, I do not believe my employer must abide to my ethical framework.

Outside of work, I do not use generative AI for coding purposes.

Exploration, practice, tinkering and enjoyment are much more valuable to me than any short-term speed gained from the use of assisted coding tool.

Beyond the practical and ethical reasons, I find LLM-assisted coding mostly boring and un-engaging. The constant context switching is unbearable.

There is real detachment from the code when you use agentic coding tool which I dislike.

I personally believe that it is often a detriment to the craft, and by association, to the end product.

I use generative AI to conduct research when I don't quite know how to formulate my questions.

I believe that the natural processing of LLMs is a true breakthrough and an engineering feat.

I do not use generative AI to summarise information and I avoid consumption of AI generated content as much as practical.

This is the reverse application of the principle that if it's not worth the writing time, it's not worth the reading time.

I avoid open source projects relying on AI-assisted coding tools as much as possible.

This is motivated both by ethical reasons, such as copyright infringement, as well as more practical reasons. I believe that the time-to-market argument of agentic coding is orthogonal to a qualitative open source project.