RAG LLM

I want my LLM to help me with my task, which is described in a two page PDF. What technique should I use ? 

• Prompt engineering: add knowledge in the prompt  

• RAG to retrieve information from the PDF 

• Fine-tune the model on the PDF 

• RAG + fine-tune 

I want to create an LLM that could answer to my clients. To train it, I have a lot of past conversations with clients and I asked my employees to produce a few high quality responses to some generic questions. What technique should I use ? 

• Prompt engineering: add knowledge in the prompt  

• RAG to retrieve information from the quality responses  

• Fine-tune the model on past conversations and quality responses afterwards 

• Fine-tuning + prompt engineering 

Where does the biases come from in AI models ? (stereotypes in LLMs) 

• It comes from the data the models are trained on  

• It comes from the own biases of the researchers that produce the LLMs 

• It comes from the pre-processing of data 

• A combination of all of the above 

What is referred as the “sequence length” in a LLM ? 

• Max number of words a LLM can take as input  

• Max number of tokens a LLM can take as input  

• Max number of words a LLM can output 

• Max number of tokens a LLM can output 

What is the field in which AI has the most potential ? 

• Write code  

• Make code reviews 

• Detect root cause issues in software services  

• Provide first level support to customers  

In the end, what does “generative” mean in generative AI ? 

• Those AI models are creative, they produce unexpected outputs  

• It refers to the auto-regression process: it predicts one word after the other 

• Well, those AI models are the only machine learning models that generate something 

• It is just a buzz word, in the end, all machine learning models generate something 

What is the current bottle neck in the development of AI ? 

• The lack of data (text, image,…) 

• The lack of experts  

• The lack of a better neural network architecture (better than transformers) 

• The lack of applications with added values