LLMs as Dictionaries Pros and Limits | Generated by AI

Home 2026.03

Question: Are LLM chatbots great dictionaries?

Answer:

LLMs are quite good dictionaries for most everyday purposes, but they are not true dictionaries and have important limitations compared to real lexicographic resources. Here’s a realistic breakdown:

What LLMs do very well (better than many traditional dictionaries in practice):

What real dictionaries still do significantly better:

Aspect Good dictionary (OED, MW, Wiktionary, 規範字典…) Current LLMs (2025)
Historical first attested use Usually very precise Frequently approximate or invented
Pronunciation (IPA) Systematic and reliable Often correct but sometimes inconsistent
Etymology depth Usually excellent Good for common words, frequently shallow or wrong for rare/old terms
Regional / dialect labeling Much more granular Improving but still quite coarse
Collocations & strong patterns Very systematic Good intuition but can hallucinate patterns
Censorship / sanitization Minimal (especially historical ones) Often refuses or heavily sanitizes offensive / taboo words
Stability of information Doesn’t change unless there’s a new edition Answer can change between sessions / model versions
Verifiability You can check the exact source / quotation No primary sources — it’s all compressed statistically

Quick rule of thumb most people use in 2025–2026:

So yes — for 90–95% of daily life usage, LLMs are excellent practical dictionaries.
But when precision, history, or verifiability really matter, they remain only an approximation of real lexicographic resources.

References:


Back Donate