What are LLMs

What they do:

• "Complete" some text
• Pretrained to gain general text and world knowledge, finetuned to give it assistant like traits

What we can do with just that:

1. Assistant or chat applications (ChatGPT)
2. Give it any amount of text and some sort of instruction and get a response with some text we want (RAG, document/information processing pipelines, classification, summarization, etc)
3. Give it some input text, and instead of having it respond with plain text, have it respond with text that is structured as JSON (Structured Outputs)
4. Give it some input text, have it respond with a valid JSON object, parse that objects values and use them as inputs to something, like a tool
5. Do #4 but over and over and over until we don't want it to anymore. (Agents!!)