What are the limitations of LLMs
LLMs) have limitations such as generating plausible but incorrect outputs (hallucinations), limited context windows, sensitivity to biased training data, and lack of true reasoning or understanding. These constraints affect their reliability and applicability in critical tasks.Large Language Models (LLMs) are AI systems that generate human-like text by predicting the next word based on vast training data but have inherent limitations in accuracy, context, and reasoning.How it works
LLMs generate text by predicting the next token based on patterns learned from massive datasets. Imagine a very advanced autocomplete that guesses what comes next in a sentence. However, unlike humans, they do not understand meaning but rely on statistical correlations, which leads to limitations like hallucinations and shallow reasoning.
Concrete example
Consider a prompt asking an LLM to provide a historical fact. The model might confidently generate a false statement because it predicts plausible text rather than verifying facts.
from openai import OpenAI
import os
client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])
response = client.chat.completions.create(
model="gpt-4o",
messages=[{"role": "user", "content": "Who was the first person to walk on Mars?"}]
)
print(response.choices[0].message.content)No human has walked on Mars yet.
When to use it
Use LLMs for tasks like drafting text, brainstorming, summarization, and coding assistance where approximate correctness is acceptable. Avoid relying on them for critical decisions, factual verification, or tasks requiring deep understanding or long-term memory beyond their context window.
Key terms
| Term | Definition |
|---|---|
| Hallucination | When an LLM generates plausible but factually incorrect or nonsensical information. |
| Context window | The maximum amount of text an LLM can consider at once, limiting long conversations or documents. |
| Bias | Systematic errors in LLM outputs caused by skewed or unrepresentative training data. |
| Token | A piece of text (word or subword) that the model processes as a unit. |
| Statistical correlation | The basis for LLM predictions, relying on patterns in data rather than true understanding. |
Key Takeaways
- LLMs predict text based on patterns, not true understanding, causing errors called hallucinations.
- Their limited context window restricts handling very long documents or conversations.
- Bias in training data can lead to unfair or inaccurate outputs.
- Use LLMs for creative or assistive tasks, not for critical factual decisions.
- Always verify important information generated by LLMs with trusted sources.