LLM (Large Language Model)
Qué es LLM (Large Language Model)?
Esta entrada está disponible actualmente solo en inglés.
A Large Language Model (LLM) is a machine-learning model trained on vast amounts of text to understand and generate human language. It is the engine behind modern conversational AI: given a customer's question in plain words, it produces a fluent, context-aware answer, which is why voice agents and support bots no longer need rigid decision trees to hold an open-ended conversation.
More detail
An LLM is built on the transformer architecture and trained to predict the next token in a sequence, across billions of examples. That simple objective, at scale, yields a model that summarizes, translates, classifies, and follows instructions. Its knowledge is frozen at training time, so production systems ground it in current, domain-specific sources at query time — the pattern known as retrieval-augmented generation, or RAG.
Putting an LLM in front of customers is an engineering problem as much as a modeling one: turn latency decides whether a voice conversation feels natural, guardrails keep answers inside what the business actually supports, tool calls let the model look up an order instead of guessing, and human-in-the-loop handoff catches the exchanges a person must own. Orchestration — choosing the model, the context, and the tools per turn — is where product quality is won.
On Devotel Orbit, LLMs power the AI voice agents and AI agents that answer calls and messages on real telephony and messaging channels, with LLM orchestration wiring the model to your knowledge base and tools and handing the conversation to a human when needed. The related terms below break out each piece of that stack.
Preguntas frecuentes
- What is the difference between an LLM and a chatbot?
- A chatbot is the interface; an LLM is one possible brain behind it. Scripted chatbots walk a decision tree and fail outside it. An LLM-based agent understands questions phrased in words nobody anticipated and answers from grounded knowledge, which is the difference between a bot that recognizes ten phrases and one that holds a conversation.
- How does an LLM handle a phone call?
- The call runs through a pipeline: speech-to-text converts the caller's audio to text, the LLM understands the request and generates a reply against your knowledge base, and text-to-speech speaks it back. The whole loop has to complete in well under a second per turn, so production voice AI budgets latency across every stage and streams audio as soon as the first words are ready.
- Why do LLM answers sometimes need retrieval or RAG?
- An LLM's parameters contain what it saw during training, not your current pricing, policies, or order states. Retrieval-augmented generation fetches the relevant passage from your knowledge base and hands it to the model with the question, so the answer cites your source of truth instead of the model's frozen guess — the standard fix for outdated or invented answers.
See also
Build it on Orbit
Voice, messaging, email, video, and AI agents on one platform and one pay-as-you-go bill. Start free — no credit card required.