意见 (Opinion)

Disclaimer: This is an opinion piece. I’d love to hear your thoughts in the comments.

免责声明：这是一个观点。 我很想听听您在评论中的想法。

Rather than ask if GPT-3 will make coders obsolete.

而不是问GPT-3是否会使编码器过时 。

Let’s assume that at some point, AI can write flawless code.

假设在某个时候，AI可以编写完美的代码。

Will there still be a place for humans writing code? Yes.

人类还会有写代码的地方吗？ 是。

编码是与AI通信的最有效方式 (Coding is the most efficient way to communicate with AI)

Code is designed to be as high-level and unambiguous as possible.

代码被设计为尽可能高层且明确。

While considered a dark art to non-developers, most coding languages are more concise than spoken languages.

尽管对非开发人员来说是一门黑手艺，但大多数编码语言都比口头语言更为简洁。

I’ll say that again. Writing out the logic of an application using English would take more words than writing it in Ruby or Python.

我再说一遍。 与使用Ruby或Python编写应用程序逻辑相比，使用英语编写应用程序逻辑要花费更多的单词。

For this reason, telling AI what to build (while navigating edge cases and domain knowledge) may be more work than writing the code.

由于这个原因，告诉AI要构建什么(在浏览边缘案例和领域知识的同时)可能比编写代码还要费劲。

For example. A simple command to an AI assistant, “Buy me toilet paper” has a lot of assumptions baked in. These could be interpreted disastrously wrong if not coded as constraints in advance. How important is price? Softness? Delivery date? Quantity?

例如。 给AI助手一个简单的命令“给我买卫生纸”有很多假设。如果不事先编码为约束，这些假设可能会造成灾难性错误。 价格有多重要？ 柔软度？ 邮寄日期？ 数量？

Coding forces an intelligent developer to consider these.

编码迫使聪明的开发人员考虑这些。

So while coding may become even higher level than it is now, it might be the most efficient way to talk to AIs.

因此，尽管编码可能变得比现在更高，但它可能是与AI对话的最有效方法。

AI编写的代码将需要进行测试(带有代码) (AI-written code will need to be tested (with code))

Given that AI could be writing code pertaining to anything, the output space is potentially infinite.

假设AI可以编写与任何事物有关的代码，那么输出空间可能是无限的。

So while you can monitor a self-driving car for 100 million miles to verify it’s safety, you can’t write tests covering an infinite space and number of domains.

因此，尽管您可以监视无人驾驶汽车1亿英里以验证其安全性，但是您无法编写涵盖无限空间和域数的测试。

This leaves us with having to test the AI-outputted code, rather than the coding mechanism itself.

这使我们不得不测试AI输出的代码，而不是编码机制本身。

As this should be approached in a logical manner, and allow retesting as applications change, it makes a lot of sense to write tests in code (at least in the beginning of AI’s development career).

因为这应该以一种逻辑的方式进行处理，并允许在应用程序更改时进行重新测试，所以用代码编写测试(至少在AI开发生涯的初期)是很有意义的。

Though at a further point in the future I can imaging another AI-layer that assists in writing tests alongside a human domain expert.

尽管在将来的某个时候，我可以对另一个AI层进行成像，以帮助与人类领域专家一起编写测试。

AI编码器可能不符合成本效益 (AI-coders may not be cost effective)

OpenAI gave prohibitive cost a reason they’ve offered GPT-3 as an API rather than an open source package.

OpenAI之所以付出高昂的成本，是因为他们提供GPT-3作为API而不是开源软件包的原因。

We’re hopeful that the API will make powerful AI systems more accessible to smaller businesses and organizations.- OpenAi

我们希望API将使小型企业和组织更容易使用强大的AI系统。- OpenAi

Given this, I don’t expect to see it as a $20/hour service on AWS anytime soon. Humans will be writing code until the price comes down.

鉴于此，我不希望很快将其视为AWS上每小时20美元的服务。 在价格下降之前，人类将一直在编写代码。

At the current moment, we don’t really know what the price is, only that OpenAI has received about $1 billion in funding.

目前，我们还真的不知道价格是多少，只是OpenAI已经获得了大约10亿美元的资金。

And while it could make sense for large dev shops to automate writing repetitive code (even at a high cost), software engineers in startups do a lot more than write code.

尽管大型开发人员可以自动执行重复代码(即使成本很高)也很合理，但新兴公司的软件工程师所做的工作远不止编写代码。

日常活动包括： (Daily activities include:)

• writing and reviewing tickets and code编写和审查票证和代码
• discussing user experience讨论用户体验
• interviewing potential hires面试潜在员工
• discussing constraints on hypothetical features…讨论对假设特征的约束...

So a software engineer’s generalist skill set could still make them a good deal compared to an AI that can only code.

因此，与只能编码的AI相比，软件工程师的通才技能仍然可以使他们受益匪浅。

That said, it’s also possible that developers become product managers, using their technical/product skills in helping manage AI’s who do the coding.

也就是说，开发人员也有可能成为他们的产品经理，利用他们的技术/产品技能来帮助管理进行编码的AI。

我们可能不相信AI与关键任务系统 (We may not trust AI with mission critical systems)

We’ll trust AI to build static Wordpress pages and “yet-another-social-media-app” applications, but will we trust it writing code for the military?

我们将信任AI来构建静态的Wordpress页面和“另一个社交媒体应用”应用程序，但是我们会信任AI为军方编写代码吗？

What is the downside of the AI being hacked, or writing faulty code?

AI被黑客入侵或编写错误代码有何弊端？

Writing flawless code within a single function is easy. Across a whole app, it’s much harder. But when it gets to the infrastructure level, it’s no longer about right and wrong, it’s about financial/business constraints and desires.

在单个函数中编写完美的代码很容易。 在整个应用程序中，这要困难得多。 但是，当涉及到基础架构级别时，它不再是非是非，而是关于财务/业务限制和期望。

We can imagine that in situations of layered complexity, a required understanding of the outside world, or significant downside, we may not want AI writing code.

我们可以想象，在复杂的层次结构，对外界的了解或明显的不利情况下，我们可能不希望AI编写代码。

自己建造技术充满乐趣 (There is joy in building technology yourself)

Long live the coding hobbyist.

编码爱好者万岁。

This is anecdotal, but I became a developer because it’s the only job I’d do for free if it wasn’t my day job.

这是轶事，但我成为开发人员是因为，如果不是我的日常工作，这是我唯一免费做的工作。

There’s a subgroup of people who enjoy building things with code for their own pleasure. It’s why people build an AI-assistant on a $100 RaspberryPi when they could just buy an Amazon Echo for $50.

有一群人喜欢为自己的乐趣而使用代码来构建事物。 这就是为什么人们可以以50美元的价格购买Amazon Echo时，在100美元的RaspberryPi上构建AI助手的原因。

Humans are craftspersons by nature and gain satisfaction from making things. This won’t be a huge group but I expect it to continue existing.

人类天生就是手Craft.io者，他们从做事中获得满足。 这将不是一个庞大的团队，但我希望它会继续存在。

丰富的经验和基础知识是创新的前提 (Deep experience and a knowledge of fundamentals are a prerequisite to innovation)

Humans need to keep coding if they hope to keep innovating in the space.

如果人们希望在太空中不断创新，就需要保持编码。

AI is great at copying what has been done. But not at combining existing concepts in new ways to create something new. We’re not talking about painting a better picture here, but developing a new type of art, or a new data transfer protocol.

AI非常擅长复制已完成的工作。 但是不能以新的方式组合现有概念来创建新事物。 我们不是在这里画更好的图片，而是在开发一种新型的艺术品或新的数据传输协议。

Most of our modern technology arose like this. From experts and dreamers frustrated with the status quo, and who knew their tools well.

我们大多数现代技术都是这样产生的。 来自专家和梦想家，对现状感到沮丧，谁对他们的工具了如指掌。

In software development, GraphQL was invented to deal with the limitations of the existing REST. The former makes frontend development easier, but didn’t “need” to build.

在软件开发中，发明了GraphQL来解决现有REST的局限性。 前者使前端开发更容易，但不需要“构建”。

Will AI learn to invent, or just do existing actions more efficiently?

人工智能会学习发明发明还是只是更有效地执行现有行动？

结论 (Conclusion)

This post was a thought experiment and based on my experience in software development, ML, and startups.

这篇文章是一个思想实验，是基于我在软件开发，机器学习和创业方面的经验。

While I may seem anti-AI, I am not. On the contrary, AI’s that can code would be the biggest opportunity for small entrepreneurs in the history of civilization because it would allow them to focus on problems instead of on the tech.

虽然我似乎是反AI的，但我不是。 相反，可以编码的AI将是小企业家在文明史上的最大机会，因为它可以使他们专注于问题而不是技术。

That said, we’re not close to that point. Despite fear mongering, we’re a long way from the rise of the robots. So while you should upgrade your skills, I wouldn’t lose sleep over GPT-3 taking your coding job.

就是说，我们还没到那个时候。 尽管恐惧四处蔓延，但距离机器人的崛起还有很长的路要走。 因此，尽管您应该提高自己的技能，但是在从事编码工作方面，我不会因为GPT-3而失眠。