Microsoft made an AI this time to let programmers leave work early and lay off early

Gpt-3是一种自然语言处理人工智能，去年"出局"。它被公认为继alphago之后"人工智能史上的又一里程碑"。它在互联网上"阅读"了5000亿字，而这个模型有1750亿个参数，这使得它具有普遍性。但是给人工智能提供硬数据是浪费金钱和精力的。但幸运的是，gpt-3正在培育可以商业化的后代。

Gpt-3, a natural language processing AI, was "out of the circle" last year. It is recognized as "another milestone in the history of artificial intelligence" after alphago. It has "read" 500 billion words on the Internet, and the model has 175 billion parameters, which makes it universal. But it's a waste of money and effort to give AI hard data. But fortunately, gpt-3 is breeding future generations that can be commercialized.

刚才，GitHub、openai和微软联合创建了新一代代码AI:GitHub copilot（以下简称copilot）。

Just now, GitHub, openai and Microsoft jointly created a new code generation AI: GitHub copilot (hereinafter referred to as copilot).

简而言之，副驾驶可以根据程序员的注释编写代码，自动完成代码，提供与代码匹配的测试，并生成多个备选代码供选择。GitHub说，copilot生成的大部分代码都是原创的。

In short, copilot can write code according to the programmer's comments, automatically complete the code, provide tests matching the code, and generate multiple alternative code for selection. GitHub says that most of the code generated by copilot is original.

你配对编程，被人工智能取代

And you pair programming, replaced by AI

在此之前，程序员不会写代码。他们通常会去stack overflow，一个软件开发领域的问答网站，以供参考，或者求助于一些代码完成工具。但是，传统的ide基本上都是用搜索的方法来完成的，也就是说有些在线，可以在过去抓到来完成。

Before that, programmers can't write code. They usually go to stack overflow, a Q & a website in the field of software development, for reference, or turn to some code completion tools. However, the traditional ide basically uses the search method to complete, in other words, some online, can be caught in the past to complete.

但副驾驶是另一种思维方式，它直接依赖于人工智能来理解程序员的需求。Copilot不仅模仿以前看到的代码，还分析字符串、注释、函数名和代码本身，以便生成新的匹配代码，包括以前调用的特定函数。

But copilot is another way of thinking, it directly depends on AI to understand the needs of programmers. Copilot not only mimics previously seen code, but also analyzes strings, annotations, function names and code itself in order to generate new matching codes, including specific functions previously called.

弗里德曼说，成百上千的GitHub开发人员在编码时整天都在使用copilot函数，他们中的大多数人接受了人工智能的建议，并没有关闭它。

Friedman said that hundreds of GitHub developers used the copilot function all day while coding, and most of them accepted the AI's advice and didn't turn it off.

但就目前而言，副驾驶建议的准确性似乎很低。GitHub对一组Python函数进行了基准测试，测试团队清除了这些函数，并要求副驾驶填写它们。第一次尝试时，模型的准确率为43%，第十次尝试后，模型的准确率为57%。该小组表示，之后，副驾驶将进行几次训练和测试，准确率将进一步提高。不过，一些程序员认为，这种自动代码生成工具必须在能够准确描述要实现的功能的前提下"易于使用"，否则后续的调试将更加耗时。

But for now, the accuracy of copilot's recommendations seems to be low. GitHub benchmark a set of Python functions, and the test team cleared the functions and asked copilot to fill them in. The accuracy of the model is 43% in the first attempt, and 57% after 10 attempts. The team said that after that, copilot will conduct several training and tests, and the accuracy rate will be further improved. However, according to some programmers, this automatic code generation tool must be "easy to use" on the premise that it can accurately describe the functions to be implemented, otherwise the subsequent debugging will be more time-consuming.

此外，GitHub指出copilot提出的代码可能无法编译或运行，copilot也可能使用旧的或废弃的库和语言。此外，副驾驶员建议的代码在试图理解开发人员的意图时可能并不总是有效或有意义的，因此程序员像往常一样仔细地测试和审查所有代码。

In addition, GitHub indicates that the code proposed by copilot may not compile or run, and copilot may also use old or abandoned libraries and languages. In addition, the code suggested by copilot may not always be valid or meaningful when trying to understand the developer's intentions, so programmers test and review all the code carefully as usual.

对于代码安全，GitHub表示，并不是所有使用的代码都经过了漏洞、不安全行为或个人数据审查。因此，他们还设置了一些过滤器，以防止副驾驶产生攻击性语言。目前的效果并不完美，需要以后改进。

For code security, GitHub said that not all the code used has been reviewed for vulnerabilities, unsafe practices or personal data. Therefore, they have also set up some filters to prevent copilot from generating offensive language. The current effect is not perfect and needs to be improved later.

但正如开放人工智能（openai）首席执行官萨姆•奥尔特曼（Sam Altman）所言，程序员的价值将下降："这是我的预测：人工智能将使那些在计算机前工作的人比那些在现实世界中工作的人更快地失去价值。这与大多数人的期望完全相反。"

But as Sam Altman, CEO of open AI, said, the value of programmers will decline: "this is my prediction: AI will make those who work in front of computers lose their value faster than those who work in the physical world. This is completely contrary to most expectations. "

有了副驾驶，程序员可以提前下班，但将来，他们也可能提前下岗。

With copilot, programmers can leave work early, but in the future, they may also be laid off early.

本文最初由geekpark（一个极客公园）发表。如果您想重印，请添加geek（ID:geekpark）

This article was originally published by geekpark, a geek park. If you want to reprint it, please add geek (ID: geekpark)

Link：https://new.qq.com/omn/20210701/20210701A00I7600.html

update time：2021-07-01 10:50:23