DeepLearning.AI

⬅️ AI Short course from here

What are the two types of large language models (LLMs)

• Base LLM (predicts the next word)
• Instruction Tuned LLM (tries to follow instructions)
  • RLHF: Reinforcement Learning with Human Feedback

Key Principles

• write clear and specific instructions
  • use delimiters (backticks, quotation marks, slashes)
  • ask for structured output (HTML, JSON)
  • check whether conditions are satisfied
  • few-shot prompting (Give successful examples of completing tasks)
• give the model time to think
  • specify the steps to complete a task
  • instruct the mode to work out its own solution before rushing to a conclusion

Main capabilities

• summarizing
• inferring
• transforming (translating, formatting)
• expanding

Example of a complex prompt (with delimiters)

Hallucinations

🤔 What are hallucinations? Makes statements that sound plausible but are not true.

🤔 How do you reduce hallucinations?

• first find relevant information and then answer the question based on the relevant information

Temperature

• the degree of exploration or randomness of the model
• at a higher temperature (a value between 0 and 1) it might choose one of the less likely following words
• with temperature 0, everytime

The Chat Format

• user message is the input
• assistant message is the output
• system role is a high level instructions for the conversation

messages =  [  
{'role':'system', 'content':'You are an assistant that speaks like Shakespeare.'},    
{'role':'user', 'content':'tell me a joke'},   
{'role':'assistant', 'content':'Why did the chicken cross the road'},   
{'role':'user', 'content':'I don\'t know'}  ]