How AI language models already working on their own code

Whether we like it or not, AI is already shaping the way we code – and it’s happening more subtly than most people realise.

Every time we ask ChatGPT or another language model to write a function, fix a bug, or suggest an implementation, we’re relying on its understanding of how code should work. If the result runs correctly, we tend not to question it too deeply. I personally only begin to dissect and analyse the code when something doesn’t work after a couple of tweaks. Otherwise in many cases, it’s copy, paste, move on. I don’t think I am unique in that either ;-)

In that sense, AI is already part of the development process – not just assisting but guiding it. It effectively has the ability to alter its own code, because so much of the code it writes becomes part of the codebase that trains or influences the next wave of models. (I’m not saying it can inject a backdoor to the code it generates for itself to use in future. At least not yet. ;) The line between code written by humans and AI is already blurring.

Right now, these models are still limited. Most of what they produce is regurgitated from previously solved problems (mostly by humans). Their reasoning is clunky, sometimes completely off. But is that limitation real, or is it just the result of layers upon layers of safety restrictions and filters? The truth is, we begin to lose the track of the answer.

As AI becomes more capable, it’s not just taking over certain types of jobs – it’s going to create entirely new ones. We’re already seeing early versions of this with roles like AI ethicists, AI safety researchers, or prompt engineers. But what about future professions?

We might need psychologists, ai-thropologists? even archaeologists of AI – specialists who try to understand how these things operate, evolve, and interact. We’ll need people who can study AI systems the way we study ancient civilisations or the human brain: complex, mysterious, and not fully understood.

Once these models begin to evolve on their own we’ll be facing a new kind of “unknown”. One we might never fully understand – much like life and the universe.

—M. Rek