"Embrace AIGC Apply ChatGPT and OpenAI API" A quick overview of the whole book and the first part of basic knowledge reading

MioChan

"Embrace AIGC Apply ChatGPT and OpenAI API" A quick overview of the whole book and the first part of basic knowledge reading [Copy link]

A few days ago, I finally received the book "Embrace AIGC Apply ChatGPT and OpenAI API". After a quick look, I found that it is a book about generative artificial intelligence and its applications. There is actually very little theoretical knowledge that is difficult to understand. It mainly introduces how to apply it. I even feel that it is a manual for using ChatGPT and OpenAI API. I have been using ChatGPT since it was launched. It is very convenient to let GPT help write nonsense or clear logic code.

The first part is almost an introduction, which mainly introduces the concept of generative artificial intelligence, its development history, and the mathematical principles behind related technologies. AIGC is simply a type of artificial intelligence model that generates new content by learning a large amount of data. For example, the first chapter also lists several common AIGC applications, such as ChatGPT, a large language model, image generation models, and music generation models. In fact, there are too many AIGC applications now. In addition to these most commonly used ones, there are also voice-oriented ones such as TTS, a voice generation model, and in addition to two-dimensional images, 3D models and video AIGC are also developing rapidly.

至于生成式AI的研究历史，书中主要列举了生成对抗网络（GAN）、变分自动编码器（VAE）、TransFormer等等，其实生成式AI再往早一点说，还有基于统计方法，如朴素贝叶斯模型和隐马尔可夫模型（HMM）这些。Transformer提出应该算是一个里程碑的事件，可以说是彻底取代了RNN，模型结构图如上所示，因为采用了自注意力机制，使得模型能够高效地捕捉序列数据中的长程依赖关系，在自然语言处理任务中表现出色，成为许多生成模型的基础。随后OpenAI提出了GPT模型，GPT也是基于Transformer架构的大型语言模型实现的，核心思想是预训练和微调。GPT-1是首个GPT模型，展示了预训练与微调结合的有效性。GPT-2的模型规模显著增加，能够生成高质量、连贯的长文本，展现了更好的生成能力。GPT-3的模型参数达到1750亿，进一步提升了生成文本的质量和多样性，能够在多个任务上实现零样本和少样本学习能力。GPT-4继续在模型规模和生成能力上取得突破，支持多模态输入（如图像和文本），在跨领域任务中展现了更广泛的应用。后续的具体原理因为OpenAI逐渐变成了CloseAI，也没有开源所以详细的细节也不得而知。

I remember that when registering an Oppo AI account, you had to verify your phone number and other things, and the steps were quite complicated, but now it seems that you don’t have to verify your phone number, and registration is still very easy. In addition, the development of LLM in China is pretty good now, so it’s worth trying if you haven’t played it. Finally, a picture of the development history of LLM is attached for your reference.

freebsder

Thanks for sharing, the “Questioner” manual