We have recently heard that GPT-3, a Natural Language Processing(NLP) models have been made available by Open AI and is touted to be an invention bigger than blockchain. The extent of this transformation is such that this deep learning model can help things shown as sci-fi to become accessible to be implemented by college students.

我们最近听说，Open AI已经提供了GPT-3(一种自然语言处理(NLP)模型)，并被吹捧为一项比区块链更大的发明。 这种转变的程度使得这种深度学习模型可以帮助科幻所示的事物变得易于被大学生实现。

Technically speaking GPT-3 is a highly adaptable algorithm that generates models but let’s not get bogged down by semantics.GPT-3 has been defined as:

从技术上讲，GPT-3是一种高度适应性的算法，可以生成模型，但不要被语义所困扰.GPT-3被定义为：

task-agnostic Natural Language Processing Model that requires manual tuning.

与任务无关的自然语言处理模型，需要手动调整。

So it can do specific tasks like text generation and qualitative queries with minimal user adjustment. Historically models have been either task agnostic or have required minimal tuning, but GPT-3 has both the properties and hence is quick to train and easy to tune.NLP is a branch of AI that deals with and enables the interaction between computers and humans using the natural language. The NLP model achieves this by reading, deciphering, understanding, and making sense of human words in the given context.

因此，它可以用最少的用户调整来完成特定的任务，例如文本生成和定性查询。 从历史上看，模型要么是与任务无关的，要么需要最小的调整，但是GPT-3具有这两种特性，因此训练Swift且易于调整.NLP是AI的一个分支，致力于处理并允许使用自然语言。 NLP模型通过在给定的上下文中阅读，解密，理解和理解人类单词来实现这一目标。

GPT-3 has been trained using half a trillion words (Common Crawl Dataset) and based on that has generated 175 Billion parameter model. Parameters are different rules that are set in different conditions by learning how language works. The complexity of the problem GPT-3 is trying to solve is demonstrated by the fact that the model consumed $12 Million worth of compute to generate this 175B parameter model.

GPT-3已使用半万亿个单词( 通用抓取数据集 )进行了训练，并在此基础上生成了1,750亿个参数模型。 参数是通过学习语言的工作原理在不同条件下设置的不同规则。 该模型耗费了价值1200万美元的计算来生成此175B参数模型，这一事实证明了GPT-3试图解决的问题的复杂性。

GPT-3 Use Cases

GPT-3用例

Following is a shortlist of a few amongst the millions of use cases

以下是数百万个用例中的几个的入围列表

1. Long-form Generative text: GPT-3 can generate essays similar to what humans would write in 10 sec. When this was tested by asking humans to guess which text was written by human and which one was generated by the algorithm, it was found that in the best case was 88% of the people were fooled to believe GPT-3 generated text was human-generated and in the worst case 39% of the people were fooled. So GPT-3 could generate text pretty close what humans would write and could even be used for creative writing like in this example => GPT-3 Creative Fiction

   较长的生成性文本： GPT-3可以生成类似于人类在10秒钟内写的文章。 当通过要求人们猜测哪个文本是由人类编写以及哪个文本是由算法生成进行测试时，发现在最佳情况下，有88％的人被愚弄了，以为GPT-3生成的文本是人类-产生，最坏的情况是39％的人被愚弄了。 因此，GPT-3可以生成与人类所写文本非常接近的文本，甚至可以用于创造性写作，例如本例=> GPT-3 Creative Fiction

2. Qualitative Text Search: We have often been frustrated when we have searched on a webpage and have been frustrated as no results have been found since the exact matching word was not used on the webpage. GPT-3 can help solve that by letting us humans describe what we are looking for and it will return a result even if it doesn’t match the word in the article as can be seen in this example video => https://cdn.openai.com/API/FINAL_bread.mp4

   定性文本搜索：当我们在网页上进行搜索时，我们经常会感到沮丧，并且由于未在网页上使用完全匹配的单词而找不到任何结果，因此感到沮丧。 GPT-3通过让我们的人类描述我们正在寻找的内容来帮助解决这一问题，即使它与文章中的单词不匹配，它也会返回结果，如本示例视频所示=> https：// cdn .openai.com / API / FINAL_bread.mp4

3. Complemet/completion tasks: GPT-3 can be used for various completion tasks like compiling data researching based on a particular context as in this example => https://cdn.openai.com/API/excel_tabulate_v3_biz.mp4

   补全/完成任务： GPT-3可用于各种补全任务，例如根据特定上下文编译数据研究，如本例=> https://cdn.openai.com/API/excel_tabulate_v3_biz.mp4

4. Code Generation: GPT-3 could also be used for useful and context-aware code generation as in the example below.

   代码生成 ：GPT-3也可以用于有用的和上下文感知的代码生成，如下例所示。

Other use cases include Chatbot, and it could also be used to generate short/long from literature like this poem on Elon Musk by Dr Seuss .

其他用例包括Chatbot，它也可以用于从文学作品中产生长短，例如苏斯博士在《埃隆·马斯克》上的这首诗 。

GPT-3 History

GPT-3历史

GPT-3 is an evolution on GPT 1, and GPT 2 and hence is an adaptation of what’s known as a Transformer, an invention by Google in 2017. The Transformer utilises a function called attention to calculate the probability that a word will appear in the given surrounding words. The code for GPT-3 will not be available like GPT-2 instead access has been provided through an API, and there is currently a waitlist for the same. The API operates in Text in Text out mode for any English language task. The API also allows you to tune performance for specific tasks by training on a dataset using a dataset provided or by learning from feedback from users or labellers. Open AI has not opened the code to ensure level playing field so that it doesn’t enable companies with big pockets to get an unfair advantage as the code is very expensive to run. Other reasons for restricting access to code are to ensure that fake news and bots are not enabled and to prevent potentially prejudiced models something which has been brought to everyone’s notice by the Black Lives Matter movement. Another reason is that the API model allows Open AI to more easily respond to the misuse of the technology and have better control over downstream use cases.

GPT-3是GPT 1和GPT 2的改进，因此是对Google于2017年发明的“变形金刚”的改编。变形金刚利用称为注意的功能来计算单词出现在“变形金刚”中的概率。给定周围的单词。 GPT-3的代码将不像GPT-2那样可用，而是通过API提供了访问权限，并且目前有一个等待列表。 对于任何英语任务，API均以“文本输出”模式运行。 该API还允许您通过使用提供的数据集对数据集进行训练，或者通过从用户或标签制作者的反馈中学习来调整特定任务的性能。 Open AI尚未开放代码以确保公平竞争，因此，由于运行代码非常昂贵，因此它不会使财大气粗的公司获得不公平的优势。 限制访问代码的其他原因是，确保不启用虚假新闻和漫游器，并防止潜在的偏见模型，这些问题已被“黑住事”运动引起了所有人的注意。 另一个原因是，API模型使Open AI可以更轻松地应对技术滥用，并更好地控制下游用例。

Thus, in conclusion, GPT-3 is a significant step forward in terms of NLP with exciting use cases ahead of us.

因此，总而言之，GPT-3在NLP方面是向前迈出的重要一步，我们面前还有激动人心的用例。

Hopefully, this article has enabled you to get a better understanding of GPT-3. For feedback or suggestions about what you would like me to write about next, please comment below or contact me on cssprashant@gmail.com.

希望本文使您能够更好地了解GPT-3。 有关您希望我接下来写些什么的反馈或建议，请在下面发表评论或通过cssprashant@gmail.com与我联系。

Refrences:-

引用：-

1. Language Models are Few-Shot Learners

   语言模型是学习者很少

2. Open AI API

   开启AI API

3. OpenAI’s gigantic GPT-3 hints at the limits of language models for AI

   OpenAI的巨大GPT-3暗示了AI语言模型的局限性

4. Elon Musk By Dr. Seuss (GPT-3)

   苏隆博士(GPT-3)的Elon Musk

5. GPT-3 Creative Fiction

   GPT-3创意小说