Meta Mania
Metaprogramming, Artificial Intelligence, doomer predications and a bunch of other buzzwords are going to be in this blog post, so, you've been warned.
Last week, GitHub(Microsoft) just released a beta version of a VS Code(Microsoft) extension called “GitHub CoPilot”. CoPilot is an AI-paired programmer that is powered by OpenAI(Sorta Microsoft). It will autocomplete and anticipate code as you start to type. From a newbie, the examples of it in action are quite impressive and concerning.
From the documentation and some live examples the essential premise is that you begin typing your code give the AI some context and then the software will try and predict what you are trying to code. The predictive CoPilot can supposedly synthesis code in a variety of high-level languages including JavaScript, Python, TypeScript, and Ruby. The reviews on the accuracy and usefulness of the suggested code vary. Some of the time it generates unique solutions the optimize what they are trying to do, other times CoPoilot completely missed the mark. The extension is still in development mode where the tested are encouraged to give feedback on the responses. I am currently on the waitlist to give it a shot but alas they have yet to grant me access.
At the core of this new VS Code extension is GPT-3, Generative Pre-Trained Transformer 3, by OpenAI. We will get into the weeds on what GPT-3 actually is in a minute but I would like to talk about OpenAI for a bit.
The company was founded, by Elon Musk and a small group of investors as a non-profit, in 2015. The companies goal was to use neural networks and deep learning models to leverage Artificial Intelligence into useful applications. Initially, the company was open source, it allowed its research and data to be freely used all while they scrapped the internet for useful data to train the modules. They would go through publiclly avaible code on Stack Overflow and GitHub. In 2018 the researchers at OpenAI published their first Generative Pre-Trained Transformer. In 2019, Microsoft invested 1 billion dollars in the company. Then in September of last year, it was announced that Microsoft had obtained the exclusive rights to use GPT-3. Doesn’t sound very open source to me. But what does Microsoft see in the data module and why did they think it was worth a billion dollars?
The GPT-3 in the latest and greatest artificial intelligence focused on producing human-like text. There are examples of the AI being used to create stories and other text-based strings that look like they were written by a human but were totally synthesized by a computer. The GPTs are considered “few-shot” learners. A full AI that has been finely tuned has the opportunity to review countless test cases that will match what they anticipate to compute with the module. A “few-shot’’ AI can be given a test that it may have seen once or twice but is able to contextualize what the user wants and give it to them in a meaningful way. This is valuable in real-world applications, allowing someone with not a lot of understanding of the underlying neural network, myself, to be able to interact and utilize it.
From a business perspective, I see the value in this technology. From a competitive capitalist stand point, yeah Microsoft good job on scooping up this technology for a billion. From a human, aspiring web developer, I am concerned. Yeah it could be immensely valuable and allow humans to talk to the AI who will just write the nitty gritty code, making people’s lives easier and more productive. But, besides all the potential movie tropes, SkyNet, Hal 2000, Terminator, is it a good thing that the AI will be so good that it will barely need any human intervention? Is it a good thing that a corporation has taken a non-profit and cornered the market? Will there be a space for Software Engineers and Web Developers in 10 years? All things that keep me up at night.
I may be green in the Software Engineer realm but I prescribe to the idea that coding languages are just tools in the tool chest. Some tools do some jobs better than others in certain applications, some are specialized, some are more general multipurpose. As languages come and go I hope that AI synthesized auto-complete coding will just become another tool in the tool chest and not the start of the end of the industry I am getting started in.
Sources/Further reading:
