Should You Learn to Code?
Date: 30.07.2024
The Case Against Learning to Code
A recent discussion has stirred controversy in the tech community. Jen-Hsun Huang, the CEO of Nvidia, made a bold statement that learning to code is becoming increasingly pointless due to the advancements in AI, particularly, Artificial General Intelligence (AGI). According to Huang, the future will see most coding done through natural language processing, making traditional programming skills obsolete. He emphasized that it is the responsibility of the tech industry to create computing technologies that do not require traditional programming. Huang stated, “It is our job to create Computing technology such that nobody has to program, and that the programming language is human. Everybody in the world is now a programmer. This is the miracle of artificial intelligence.”
Provocative as it may sound, Huang’s assertion holds weight when considering the advancements in AI coding models like AlphaCode, which outperformed 87% of human programmers in its initial version, with expectations that subsequent versions will further improve. This raises the question, if AI can code better than humans, should we still learn to code?
Notable Advancements in Coding AIs
The landscape of coding is evolving quickly due to advancements in AI. For instance, the Gemini 1.5 Pro release introduced AlphaCode 2, which demonstrated significant improvements over its predecessors. A notable figure from the demonstration showed that Gemini 1.5 could solve about 75% of programming challenges on the first attempt, compared to 45% with earlier Palm models. If allowed to check and repair answers, this figure leaps to over 90%.
Additionally, Devon, an AI software engineer built on GPT-4, has achieved promising, albeit controversial, results. Developers at Devon have shown that by implementing a specific framework around GPT-4, significant improvements were observed in performance benchmarks. They boast a 13.86% improvement over unassisted models, highlighting the power of leveraging existing AI models to enhance coding capabilities. It’s noteworthy, though, that these results came under scrutiny. It’s clear that, despite its potential, Devon and similar systems are in early stages, and achieving reliable large-scale deployment remains a challenge.
Layoffs and Market Changes in Tech
Recent trends indicate a shake-up in the tech industry’s employment landscape. Major tech companies, including Tesla, Amazon, and Microsoft, are planning continued layoffs into 2024. This signals a shift from the high-growth phase of tech employment seen in previous years. Additionally, the influx of generative AI technologies is contributing to an increasingly saturated job market.
A tweet highlighted the current state of the industry, noting that more than 12,000 applicants vied for a $20 an hour software engineering internship, underscoring the competition and oversupply in the market. Moreover, notable AI investments like the $25 million raised by Devon and Magic’s $100 million investment illustrate the rapid influx of capital into AI, possibly at the cost of human programmers.
Advancements in Coding AIs
AlphaCode Improvements
One of the notable advancements in AI-driven coding is the development of AlphaCode, a system introduced by Google DeepMind. Initially, AlphaCode 2 could code better than 87% of human programmers. This was a significant leap in the capacity of AI to handle complex and competitive programming tasks. The release of Gemini 1.5 further showcased these capabilities, as it could solve about 75% of Python programming functions on the first try—a substantial increase from previous performance benchmarks.
To illustrate the progression, AlphaCode 1 performed at about 50% of the average human programmer’s capabilities. With AlphaCode 2 reaching 85%, the question arises: how much better will AlphaCode 3 be? As the technology continues to improve, we can expect a near-inevitable shift towards increasingly automated coding environments.
However, it is essential to consider the current limitations. According to a research paper on AlphaCode 2, the operating cost and the requirement for extensive trial and error make it impractical for large-scale applications right now. These barriers imply that while the technology is promising, a fully autonomous coding system is still not immediately on the horizon.
Deon: The First AI Software Engineer
Another groundbreaking development in AI-driven coding is Deon, hailed as the first AI software engineer. Deon is essentially a sophisticated application of GPT-4, and preliminary evaluations showed an improvement of up to 13.86% in coding efficiency. The company claims this leap is due to a framework that optimizes GPT-4’s capabilities.
However, scrutiny has revealed some inconsistencies. Deon’s much-publicized demo, which allegedly completed a task in 24 hours that an average engineer could finish in one to two hours, was found to involve some exaggerations. Additionally, issues such as the code being hard to maintain and the task completion speed being misleading cast doubts on the actual capabilities of Deon in replacing human programmers entirely.
Despite these limitations, the evolution of tools like Deon signifies a trend towards increasingly sophisticated AI applications in software development. This underscores the importance of developers adapting to integrate these AI tools into their workflows effectively.
High Stakes in AI Investments
Investment in AI technologies for coding has been ramping up significantly. For example, Deon secured a $25 million funding round shortly after its initial demo. Such investments highlight the enormous potential seen in AI-driven software engineering tools.
Additionally, companies like Magic are drawing considerable attention. CEO Nat Friedman was reportedly so impressed by a demo from this nascent company that he invested $100 million, indicating high stakes and high expectations from these emerging technologies. Although the demo details are still under wraps, Magic’s example points to a competitive and rapidly evolving landscape in AI-driven coding solutions.
The influx of capital suggests that we are on the cusp of monumental changes in software development, driven by better and more intelligent AI tools. This heightened investment environment indicates that the industry is moving quickly, and those involved must stay updated on these technological advancements to remain relevant.
Mindgine recognizes the dynamic shift in the technology landscape and aims to provide insights and consulting services to help businesses and professionals navigate the evolving AI-driven world. Stay informed with cutting-edge developments and position yourself ahead in this rapidly changing industry.
Reasons to Continue Learning Coding
Limitations of Current AI Systems
Despite the promising advances in AI, there are several **limitations** with current AI systems like AlphaCode 2. For instance, while AlphaCode 2 was better than **85%** of human competitors, it still required **a lot of trial and error** and remains too costly to operate at scale (2024). Google Research states: “A lot more remains to be done before we see systems that can reliably reach the performance of the best human coders.” This means that generating high-quality, reliable code **at scale** is still a significant challenge, making human coders indispensable for now.
Exaggerations and Misleading Demos
The high-profile launch of Devon, claimed to be the first AI software engineer, raised eyebrows with its **impressive benchmarks**. However, the claims have come under scrutiny. The Now You Know video highlights several **fabrications and exaggerations** within Devon’s capabilities. For example, the demo was described as completing tasks in **24 hours**, which would typically take a human coder only one or two hours to complete. This points to a prevalent issue in AI marketing **hype**, where system capabilities are often overstated to attract attention and **investment**.
The Value of Coding Skills in AI and Machine Learning
Continued learning in coding remains valuable as it transcends to AI and **machine learning** fields. Despite AI making strides, coding serves as the **foundation** for many AI tools and Frameworks, such as TensorFlow and PyTorch, which are built on languages like Python. Understanding coding allows professionals to utilize, customize, and troubleshoot **AI tools** effectively. Coding also provides a deeper grasp of the inner workings of **AI algorithms**, crucial for designing effective AI solutions.
As we navigate the evolving landscape of AI and automation, keeping abreast of industry changes is essential. To stay ahead, explore courses at **Mindgine Academy** where you can equip yourself with cutting-edge AI and coding skills [Mindgine Academy Course](https://academy.mindgine.com).