Lecture1 Introduction
Two large language models (LLMs, Large Language Models):
- Base LLM:
Predicts next word, based on text training data - Instruction Tuned LLM:
Tries to follow instructions
Lecture2 Guide
Two principles of using Chatgpt:
-
Write clear and specific instructions
-
Use a clear delimiter (not specified) to separate the content that needs to be summarized to prevent the wrong execution of the suspected instruction text in the pending text
-
Given a small number of training samples
-
Give the model enough time to think
-
Specify the steps required to complete a task
-
Instructs the model to think about the solution before jumping to conclusions
Models Limitations:
Lecture3 iterations
Lecture4 Summary
- The model can be used to summarize long texts to help better read the core information quickly, that is, to generate summaries
Lecture5 reasoning
Lecture6 conversion (translation)
- Can be used for grammar proofreading of documents, etc.
Lecture7 extension
-
Model Temperature selection:
-
It can be understood as the tone of the response set by Bing. More precise means that the Temperature is closer to 0, which is also the default value of Chatgpt3.5, but obviously the default value of Bing is not 0
Lecture8 Chatbot
Lecture9 Summary
- Finally,
this note is only a partial summary of the course. It is recommended to watch the original video of DeepLearning.AI.