LLM Knowledge Limitations

# Illustrating LLM knowledge limitations

knowledge_limitations = {
    "Training cutoff": {
        "problem": "The model only knows about events before its training data was collected",
        "example": "If you ask GPT-4 about news from last month, it will not know",
        "rag_solution": "Retrieve recent news articles and include them in the context"
    },
    "Private data": {
        "problem": "The model has never seen your internal company documents",
        "example": "Asking about your company's refund policy gives a generic answer",
        "rag_solution": "Retrieve your actual refund policy document and include it"
    },
    "Hallucination": {
        "problem": "The model may invent specific details it does not actually know",
        "example": "The model invents a citation or product specification that does not exist",
        "rag_solution": "Ground answers in retrieved documents so the model has actual text to reference"
    },
    "Context limit": {
        "problem": "You cannot paste an entire document library into a single prompt",
        "example": "You have 500 product manuals you want the AI to know about",
        "rag_solution": "Only retrieve the most relevant sections at query time"
    },
}

print("LLM Knowledge Limitations and RAG Solutions:")
print()
for limitation, details in knowledge_limitations.items():
    print(f"Limitation: {limitation}")
    print(f"  Problem: {details['problem']}")
    print(f"  Example: {details['example']}")
    print(f"  RAG solves this by: {details['rag_solution']}")
    print()