Large Language Model (LLM)
A type of AI model trained on vast amounts of text data that can generate, summarise, translate, and reason about natural language — powering tools like ChatGPT, Claude, and Gemini.
What Is a Large Language Model?
A large language model (LLM) is a type of artificial intelligence model trained on enormous datasets of text — web pages, books, code, articles — to understand and generate human language. LLMs work by predicting the most probable next token (word or word fragment) given a sequence of input tokens, a process refined through feedback training to produce coherent, helpful, and accurate responses.
GPT-4 (OpenAI), Claude (Anthropic), Gemini (Google), Llama (Meta), and Mistral are all LLMs. They power products like ChatGPT, Microsoft Copilot, and Notion AI.
How LLMs Work (Simply)
LLMs are trained in two stages:
-
Pre-training: The model is exposed to billions of examples of text and learns statistical patterns in language — grammar, facts, reasoning, code, and more.
-
Fine-tuning / RLHF: The pre-trained model is further refined using human feedback (Reinforcement Learning from Human Feedback) to follow instructions, be helpful, and avoid harmful outputs.
When you send a prompt, the model generates a response token by token, each choice influenced by the preceding context.
LLM Capabilities
- Text generation: Writing, summarisation, translation, paraphrasing
- Reasoning and analysis: Answering questions, explaining concepts, drawing inferences
- Code generation: Writing, reviewing, and debugging code
- Conversation: Multi-turn dialogue that maintains context
- Instruction following: Completing structured tasks from natural language descriptions
LLM Limitations
- Hallucinations: LLMs sometimes generate plausible-sounding but false information with confidence
- Knowledge cutoff: Training data has a cutoff date; LLMs don't know about recent events unless connected to live tools
- Context window limits: LLMs can only process a limited amount of text at once
- No inherent memory: Without external memory tools, LLMs don't remember previous conversations
Security Implications
LLMs present new attack surfaces: prompt injection (manipulating model behaviour through inputs), data exfiltration via AI agents, and model inversion (extracting training data). Organisations deploying LLMs must assess these risks as part of their AI security posture.