Definition
A large language model (LLM) is a neural network with billions of parameters trained on massive datasets of text from the internet, books, code, and other written sources. Through this training, LLMs learn statistical patterns in language that enable them to generate coherent text, answer questions, summarize documents, translate languages, write code, and perform reasoning tasks. The "large" in LLM refers to both the scale of the model (billions of parameters) and the volume of training data (trillions of tokens).
LLMs work by predicting the most likely next token (word or sub-word) given the preceding context. Despite this seemingly simple mechanism, the scale of training produces emergent capabilities that go far beyond simple text completion, including instruction following, few-shot learning, and multi-step reasoning. Models like GPT-4, Claude, Gemini, and Llama are the current standard.
Why It Matters for Product Managers
LLMs are the most widely adopted AI technology in product development. Product managers must understand LLM capabilities, limitations, and costs to make informed decisions about AI feature development. Knowing when an LLM is the right solution versus when simpler approaches suffice, understanding the trade-offs between different models, and being able to evaluate AI feature quality are all becoming core PM competencies.
From a strategic perspective, LLMs are reshaping competitive dynamics across every product category. Products that effectively integrate LLM capabilities can meaningfully improve user experiences in areas like search, onboarding, support, and content creation. PMs who understand the technology can identify high-impact use cases, set realistic timelines, and avoid the common trap of overpromising what AI can deliver.
How It Works in Practice
Common Pitfalls
Related Concepts
LLMs are a category of Foundation Model whose behavior is shaped by the Context Window (how much text they process at once) and Temperature (how deterministic their outputs are). Fine-Tuning specializes an LLM for a particular domain, while Hallucination remains the key reliability risk teams must mitigate in production.