Large Language Model (LLM)
An AI trained on enormous amounts of text to predict the next piece of writing — the technology behind chatbots like ChatGPT and Claude.
When not to use it
- Anything requiring a guaranteed-correct answer. An LLM produces plausible text, not verified fact. For arithmetic, lookups, or policy decisions, use a calculator, a database, or a rule — and let the model call it.
- High-volume, narrow classification. A small fine-tuned classifier will be cheaper by orders of magnitude, faster, and more accurate at telling spam from not-spam.
- Where the input is confidential and you can't control where it goes. This is a procurement question, not a technical one, and it kills more projects than any benchmark.
Reach for something else instead
- Rules and regexes are unglamorous and still correct for structured, predictable input. If a regex solves it, a regex solves it.
- Smaller task-specific models beat general LLMs on narrow jobs at a fraction of the cost.
- Traditional ML (gradient boosting and friends) remains the right tool for tabular prediction, where LLMs are simply the wrong shape.
Sources & further reading
- Brown et al. (2020), Language Models are Few-Shot Learners — GPT-3, and the demonstration that scale alone changes what models can do.
- Kaplan et al. (2020), Scaling Laws for Neural Language Models, and Hoffmann et al. (2022), Training Compute-Optimal Large Language Models — the second corrected the first on how to spend a compute budget.
- Ouyang et al. (2022), Training language models to follow instructions with human feedback — InstructGPT, the step that turned a text predictor into something usable.
Primary sources, listed so you can check the claims on this page rather than take them on trust.
Where people go wrong
- Treating fluency as accuracy. The model's confidence is a property of its writing style, not its knowledge.
- Expecting reasoning to be reliable because it's usually reliable. The failure mode is silent and looks identical to success.
- Building on a single model with no evaluation harness. Without a way to measure quality, every prompt change is a guess and every upgrade is a gamble.
At a glance
FieldLanguage & LLMs
Core mechanismnext-token prediction
Built onthe transformer
Strengthlanguage & reasoning
Weaknessfacts & reliability
DifficultyBeginner → Intermediate
Flashcards for this concept
Question
—
Answer
—
1 / 4
Often compared with
LLM vs. Traditional Search — generation from patterns vs. retrieval of existing pages.