Chatgpt bot discord how to#

There are many different databases today with vector storage support, such as Pinecone and Chroma. But how do we store and compare these embeddings? We use a database with vector support! Setting up a vector database If we split a large documentation into smaller 'chunks', we can turn them into embeddings and use them as context for ChatGPT. These vectors can be used to measure the relatedness of text strings, and we'll be using them to cherry pick context from our online documentation later on. The idea behind embeddings is that words that are semantically similar (i.e., have similar meanings) should be close together in this space.įor example, the word 'dog' might be represented as a vector with high values for features like 'four-legged,' 'bark,' and 'loyal,' while the word 'cat' might have high values for features like 'furry,' 'purr,' and 'independent'. So what are embeddings? Embeddings are a way to represent words or phrases as numerical vectors in a high-dimensional space. We will also turn the application into a Discord bot and push it to the machine learning community called Replicate as a first-line support agent.

Chatgpt bot discord how to#

In this blog post we'll learn how to use 'embeddings' to limit and cherry pick context from an online documentation page, and make ChatGPT answer questions about it that it couldn't before. ChatGPT (3.5-turbo) also currently has a context limit of 4096 tokens (or approximately 3000 English words) which means you must cherry pick your context for the instruction at hand. It's possible to 'pre-feed' ChatGPT with your own data, giving it context before you instruct it, but the context length can quickly become costly.

Altough ChatGPT may 'know' a lot about topics it was trained on, it performs poorly on topics it was not trained on, such as your own personal life, your private emails or your own documentation. It's remarkable how vast the knowledge seems to be. You can use it to debug your code, explain complicated topics in simpler terms or even improve your blog post texts. ChatGPT is a 'Large Language Model' (LLM) which is impressively good at answering questions.