Listen to this story
Can a programming language ever get replaced? One might say that C was replaced by C++, which was later replaced by Python. Though Python has been able to stand its ground amid rapidly-rising languages like Rust, the latest competitor is our beloved English — riding on prompt engineering.
Promptgramming is something that Python folks are forced to hate. As Aleksa Gordić, former DeepMind and ex-Microsoft researcher put it, “The history of computer science is the history of smirking at a thing that’s about to ‘replace’ you.”
During the Microsoft Build 2023, there was great emphasis on the fact that, “now everyone’s a developer”, with the rise of models that can write codes for everyone with just a single prompt. But on the other hand, the creator of models such as ChatGPT, Sam Altman believes otherwise.
He believes that the current state of prompt engineering is only because of the temporary limitations and capabilities of large language models. He believes in his mind that anyone who does prompt engineering right now would not be doing it in the next five years. That is not because it will become a fad, it is because everyone, including the LLMs, would be too proficient in understanding what the user wants, and generating it perfectly.
Is it a tall claim?
Firstly, it might be true technologically. Altman being the creator of such models might be able to estimate the possibilities that these LLMs hold in the future. Secondly, language being the most important tool in the world is not something that anyone would deny.
“Text is the projection of the world,” said OpenAI’s co-founder Ilya Sutskever. While he was talking about the capabilities of LLMs and that text-based models would possibly lead to AGI, it can also be said that the perfect arrangement of words to bring out meaning and information is always going to be mightier than any sword.
Coming back to programming languages. Just as Python could not replace C++, prompts may not be able to replace coding. Some people still use Assembly, C, or C++ for building frameworks and algorithms. Python is good for machine learning, and prompt engineering is good for building models quickly, but to build a foundation from scratch, C or C++ is always going to be there.
That is not to say that prompt engineering holds no value. It has already found its niche and is probably going to stay here for a while. In certain cases, a lot of prompt engineering jobs pay more than Python. Temporarily, it might look like “promptgramming” is actually replacing Python.
What will replace prompt engineering?
If we think about it, prompting is very similar to a lot of other languages. It is just a language that feels closer to human language. Going on ChatGPT or Codex and just typing in what you want does not work that easily. It’s a skill to learn as well. Anyone who has tried building models by throwing prompts at these models would know that it’s not a piece of cake.
The best part about prompting is that it is mostly just trial and error — if one prompt doesn’t work, try another. It might work out eventually. But isn’t that the case with a lot of programming languages as well? You indeed have to have a basic knowledge of coding before even jumping into Python, prompt engineering just removes that barrier to entry as well.
Apart from the legal and ethical issues about copying code from Stack Overflow and other programmers, the problem with these code generation platforms is that they are not perfect. Even if two people generate the same prompt, the results might be different. Moreover, debugging the code written by these models is a task in itself — one that would require knowledge of programming languages.
So what would kill prompt engineering? Probably in a few years as LLMs get better, it would be in everyone’s ability to generate perfect outputs every time. By the way, the most interesting thing that would kill prompt engineering would be thought prompting. With NeuraLink and similar projects coming into the picture, soon, just like Python developers smirk at prompt engineering, prompt folks would jeer at them, “These thought prompters can’t even type their own prompts. Losers!”