You Can Build Your Own Custom AI

You can build your own custom AIs in ChatGPT-4 that use your instructions, your data, and your actions. These are called GPT:

• You write in plain English. No code. (You can write in other languages).
• ChatGPT’s AI walks you through the steps to make one.
• You can make a basic GPT in ten to fifteen minutes. (By now, I can make one in less than seven minutes).
• Your GPT can create images. It can browse the web to look for more information.
• It can analyze mathematics, spreadsheets, and computer code.
• You can add your files and documents (text, PDFs, spreadsheets, etc.) for it to use as a reference. For example, you can upload your company’s instruction manuals so people can ask your GPT how to use your devices.
• Your GPT works in English, French, Spanish, Chinese, and many other languages. People can type in their language, and it will reply in that language.

Just as we build websites, you can build GPT for yourself, your friends, your team, your company, your clients, or the public.

Add Resource Augmented Generation (RAG)

OpenAI ChatGPT, Bing Chat, Anthropic Claude, Google Bard, Perplexity, and other Large Language Models (LLM) are a type of AI (there are 34+ major categories for AI). These LLM AI use a dataset, which is a collection of documents by EleutherAI, called The Pile, which has a lot of information) (okay, the equivalent of 700 million books) to generate replies. The Pile covers medicine, law, mathematics, literature, patents, philosophy, and so on. This means the LLMs are generative AI (genAI): they generate replies. Another way to say that is LLMs make up stuff.

If you want an AI that doesn't make up too much stuff, give the AI precisely the information it needs.

For example, you want an AI that answers questions on how to pass the California driving test. To do this, you give the AI the State of California Driving Test (100-page PDF, available at dmv.ca.gov/portal/file/california-driver-handbook-pdf). You then ask it questions and it answers. This is RAG in action.

RAG is a simple idea.

• You give the AI the documents it should use as the primary source of information.
• The AI then uses its genAI ability to fill in gaps or rewrite the reply for tone.
• If you use several documents, genAI combines those to create a fuller reply.

A cool thing about an LLM AI is its ability to figure out just about any kind of file: Word, PowerPoint, PDF, spreadsheets, tables, graphs, images, audio, videos, and so on. Just throw it all in the AI and let it figure it out.

Another advantage of RAG is less hallucination. Because the AI is focused on your text, it doesn’t become misled by the massive amount of information in the Pile.

The custom GPT lets you upload your files and documents for RAG. You'll be able to focus on your data.

To learn how to do this, get my free ebook at andreas.com/book-gpt.html It walks you through the steps to make a basic GPT and an advanced GPT with RAG.

-- andreas