Stay Calm About the AI Doomsday—Artificial Intelligences Can’t Write Code Yet

The Reality of AI Coding: A Cautionary Tale

In the past week, I found myself in a bit of a unique predicament—one that involved navigating the complex world of two different AI systems: ChatGPT and CLAUDE. What initially began as a straightforward approach to creating graphic files for my Substack account quickly escalated into an unexpected adventure that I would like to share.

My original plan was to utilize PowerPoint, a software I am quite comfortable with, to streamline my graphic design process. However, in a burst of enthusiasm, I decided to embark on a more ambitious route and explore what AI coding could offer. I thought, “Why not give it a shot?”

What I envisioned as a time-saving shortcut turned out to be anything but. My collaboration with CLAUDE took us on a chaotic path filled with unforeseen challenges. Every time we tackled one issue, it felt as if a whole new set of complications emerged, akin to a game of Whac-A-Mole. Without diving into the multitude of setbacks we encountered, I can candidly say that “clusterf**k” appropriately describes the situation. After days of navigating this digital labyrinth, I ultimately returned to my initial plan: a well-crafted PowerPoint presentation completed in just over three hours.

The experience served as a stark reminder that the coding capabilities of current AI systems can be likened to those of a novice coder on their very first project. Yes, advancements such as Reinforcement Learning from Human Feedback (RLHF) and AI-generated feedback will undoubtedly enhance their coding efficacy in the future. Expert human input will play a crucial role in developing smarter coding assistants.

However, for now, the existential dread surrounding artificial intelligence seems to stem more from misplaced fears or sensationalist narratives propagated by those seeking to capitalize on the technology. Instead of succumbing to anxiety about AI’s potential threats, I recommend a more relaxed approach—pour yourself a glass of Scotch, take two aspirin, or consider cannabis if that suits you better. Let’s take a step back and remember that, at this stage, AI coding may not be as formidable as some might believe.