LLM Limitations
Understanding what LLMs cannot do is as important as understanding what they can do. This helps you set appropriate expectations and design reliable systems.
8 min•By Priygop Team•Updated 2026
Core Limitations
- No real-world knowledge cutoff: training data has a cutoff date. The model does not know about recent events
- Cannot browse the internet (unless given a tool): by default, the model only knows what was in its training data
- Cannot do reliable arithmetic: LLMs make arithmetic errors, especially with large or unusual numbers. Use a calculator
- No persistent memory: each conversation starts fresh unless you explicitly include previous conversations in the context
- Context window limits: very long documents cannot be processed in one shot without chunking
- Cannot take actions: LLMs generate text. They cannot send emails, access files, or control systems unless given specific tools
Limitations in Code
Limitations in Code
# Demonstrating common LLM limitations
limitations = [
{
"limitation": "Arithmetic",
"what_LLM_might_say": "127 x 349 = 44,123",
"actual_answer": str(127 * 349),
"fix": "Use Python or a calculator for any calculation that matters",
},
{
"limitation": "Recent events",
"what_LLM_might_say": "The latest iPhone model is the iPhone 15.",
"actual_answer": "Depends on today's date. Model training data has a cutoff.",
"fix": "Use a search-enabled model or provide the current date in your prompt",
},
{
"limitation": "Memory across sessions",
"what_LLM_might_say": "[Yesterday] User: My name is Alex. [New session] User: What's my name? LLM: I don't know.",
"actual_answer": "The model has no memory of previous sessions.",
"fix": "Store user information in a database and include it in the system prompt",
},
]
print("Common LLM Limitations and Fixes:")
print()
for item in limitations:
print(f"Limitation: {item['limitation']}")
print(f" What LLM might say: {item['what_LLM_might_say']}")
print(f" Actual answer: {item['actual_answer']}")
print(f" Fix: {item['fix']}")
print()Diagram
Loading diagram…
Deep Learning ⊂ Machine Learning ⊂ Artificial Intelligence
Key Takeaways
- Understanding what LLMs cannot do is as important as understanding what they can do.
- No real-world knowledge cutoff: training data has a cutoff date. The model does not know about recent events
- Cannot browse the internet (unless given a tool): by default, the model only knows what was in its training data
- Cannot do reliable arithmetic: LLMs make arithmetic errors, especially with large or unusual numbers. Use a calculator