Microsoft Corporation at the Microsoft Build developer conference, which took place a few days ago, presented an interesting novelty. It is OpenAI's GPT-3 natural language processing model that allows you to create applications with minimal knowledge of programming languages.
Yes, there is a wide scope for jokes here, but this development has good prospects. The model converts requests in a regular language - so far only English - into code. The result of the transformation is not too complicated, yet the neural network is not able to write several thousand lines of code on a couple of common questions. But the development has potential, and considerable.
Plain text to code? Does it really happen?
In general, yes, the neural network is capable of transforming not the most difficult questions like "find products whose name begins with the word kids." The system processes such a request without any problems, giving a result like "Filter ('BC Orders' Left ('Product Name', 4) =" Kids ")".
So far, the neural network works with only one PL - the open source Power FX language based on Excel. As mentioned above, the system recognizes only one language - English. The developers promise that there will be more languages in the future.
In difficult cases, the user of such a system needs at least minimal knowledge of programming languages. The task of the platform is not to create a perfectly working program with several tens of thousands of lines of code, but to select optimal formulas that perform the required task.
Microsoft commented on the new development as follows: “Now there is a huge demand for digital solutions, but there are not enough programmers. The shortage of specialists in the United States alone is about 1 million people. Why not teach the development environment to speak the common language of people? "
Well, now - the details.
At Microsoft, the idea for such a project came about while working on the Power Platform, a specialized platform designed for corporate clients. The platform allowed and allows solving basic tasks of the corporation's partners without involving professional developers. The Power Platform can be extremely useful for companies that cannot hire experienced programmers for some reason. You can solve such tasks as basic analytics, data visualization, workflow automation.
The platform's capabilities were shown, in particular, using the example of an application created for Coca-Cola. This application monitors the supply of cola concentrate to various sites. The application interface is created in a visual designer that resembles a Power Point. However, to create a menu that allows users to interact with the database, you need a basic knowledge of Microsoft Power Fx formulas.
And here the average user gets a problem - after all, not everyone understands these formulas. It is clear that a professional programmer will do everything quickly, but, as mentioned above, not all companies can find a developer on time, and someone simply does not have money for a programmer.
In general, in order to make life easier for its customers, Microsoft Corporation has developed an algorithm that "listens" to the user, "understands" him and makes the necessary formula. So, for example, instead of the user searching the database with the query "FirstN (Sort (Search ('BC Orders'," Super_Fizzy "," aib_productname ")," Purchase Date', Descending), 10) ", he could simply write “Show 10 orders with Super Fizzy in their name and sort them by date of purchase with newest at the top,” and GPT-3 will return the correct code.
Not to say that the algorithm is super-complex or ultra-modern, but it is something that can save millions of users time and money. So far, the system is being tested, it will become available to users starting in June (preview version). Overall, Microsoft is not the first company to come up with this idea. But its implementation is extremely simple and effective. In order to reduce the likelihood of error, the user is given the opportunity to correct the final formula. In addition to reducing the number of errors, it is also the possibility of gradual user training - sooner or later, even a person far from IT will begin to understand what's what.
Technology Perspective
Microsoft has now developed a secure implementation of GPT-3. Safe in the sense that if the user makes a mistake (or the algorithm itself makes a mistake), it does not threaten anyone. The error will most likely be found and corrected.
But the implementation of other services based on GPT-3 can be somewhat dangerous. For example, a medical service, an experimental chatbot that communicates with patients and gives them medical advice, once in the course of a conversation recommended that one of the interlocutors kill himself. And it's not a joke. It is good that the service was only being tested - after all, such advice, which will be seen by a seriously ill person, can be understood and taken literally.
In order to reduce the number of potential problems, Microsoft Corporation has taken action. In particular, a list of forbidden words and phrases has been created that the system does not respond to. “If you try to get AI to generate code to implement a bad idea, you will fail. If suddenly the service fails to implement the request by mistake, you can always contact the support service, ”said one of the platform developers. Bad ideas mean something like filtering users by skin color, religion or nationality.
Overall, this is just the beginning of Microsoft's ambitious plans. In the near future, the generation of formulas on demand by voice may be included in the functionality of Excel. So far, we can talk about the introduction, in full or in part, of a new platform in products that need to work with Power Fx. In the future, perhaps the platform will become even more functional, so that it will be possible to develop simple applications simply by dictating text.