Dead Internet Theory

Anchorage in Summer...

The fear of AI replacing software engineers comes a lot across our discussions. The issue I see is more of a self-fulfilling prophecy. Students regressing to "vibe coders" and using shortcuts, rather than getting fundamentals. Such behaviors could make them easily replaceable by AI or more likely by those who don't even have a CS degree. An enthusiastic manager may decide to "vibe code" themselves, rather than hiring a fresh CS graduate to vibe code for them.

On the other hand, if the CS graduate is very talented and knowledgeable in their fundamentals and latest technologies, they may become the ones who build the AI/LLM tools. After all, we still need CS folks to build those AI tools. Those tools don't build themselves. At least not yet.

Another argument that favors CS degrees are, at least for now, LLMs are good at individual programs - but they are not sufficiently sophisticated to configure and deploy complex systems. You can write some Python (or Java, Erlang, Go, ...) code with LLM. But it is still not possible to build a hybrid cloud architecture with load balancing and security policies configured. I can compare it with dishwashers. Dishwashers may wash the dishes - but you still need to do the initial cleaning (don't throw dishes in with huge chunks of food waste in them), loading, and unloading. These tools may have made our life easier - but did not eliminate house work completely (sadly).

Similarly, the AI/LLM tools may help eliminate redundant, repetitive, and boring tasks. But they won't replace the software engineers completely. But, if our undergraduates let the AI/LLM replace them, it will replace them (individually, not collectively). I emphasize in all my courses that students should see the AI/LLM tools as an extension to themselves, rather than a replacement to themselves.

Sadly, some students tend to misuse AI/LLM in places where it won't even function properly (for example, to summarize videos; ChatGPT cannot even watch a video!)

Another point is, coding is not the only job of a software engineer. It may be just 10 - 50% of the time. Rest of the time goes with attending meetings, making presentations, design decisions, testing, ... These cannot be replaced by vibe coding after all...

I also wonder... if humans stopped making content (blog posts, videos, drawings, audio recordings, ...) AI will continue to train on the slop it itself produced and keep regurgitating recycled slop. YouTube comment sections, Twitter feeds, LinkedIn comments, even YouTube videos themselves are AI slop. We are getting close to the dead Internet theory. Hopefully, it is just a minor, temporary phase.