Nvidia’s CEO Jensen Huang recently said that, with AI, everyone could now be a programmer. Does that mean that children will no longer have to learn a programming language in the future? This is yet another topic on which we don’t feel obliged to agree with the mainstream, trendy view heard in the media.
Some people keep saying that LLMs are about to replace their own creators. If this has to happen, it’s probably quite remote. According to a study published jointly by researchers in Chinese, Australian and New Zealander universities (1), code written by GitHub Copilot, for example, is riddled with problems. On average, 35% of the code contains security issues, and this figure rises to 45% in programming languages such as C++ or GO. The problem with algorithms that learn from all available sources is that they can’t differentiate good from bad programming… and that’s without even mentioning the numerous problems linked to copyright or the impossibility of protecting AI-generated code.
Professionals didn’t wait until AI became popular to code without the complexity of indentation. For over five years now, no-code or low-code languages have been common, and are used for both learning and prototyping. Yet more and more developers are necessary, and their number grew another 3% in 2023 in the US.
What the current theory misses is that problem-solving has always been the main value added by developers. The discipline imposed by programming languages is a key skill valued throughout all industries. A developer doesn’t just produce code that does what it is meant to do. Doing better with less, or with a more astute approach, is the magic of good programming. With the way they are currently trained, AI models cannot be innovative because they don’t create code: they only copy it. AIs will be great assistants or code monkeys but, at this stage, they can’t replace a good developer’s ability to solve problems. Perfect code Is not just for glory: it minimizes storage requirements and maximizes execution speed with, in both cases, an impact on usability, operating costs and power consumption.
Innovation doesn’t always push initial players out. It may also make their skills more important, and we trust that this is what AI will do to good developers. After all, personal computing did not kill typing, it just turned all of us into unwitting typists!
AR – April 17, 2024
(1) Yujia Fu, Peng Liang, Amjed Tahir, Zengyang Li, Motjaba Shahin and Jiaxin Yu, “Security Weaknesses of Copilot Generated Code in GitHub”