Try XGEN free for 15 days — nothing to install, runs in your browserStart free trial
PlateerAI Labs
Insight
Series

LLM Inside

From next-token prediction through parameters and training, context, reasoning, and tool use. We break the workings of an LLM into core concepts and explain each in plain terms — how an answer gets made, what actually drives quality, and how far you can take it.

5 articles
  1. 1
    Tech Note

    If an LLM only predicts the next token, how does it get work done? (Part 1)

    An LLM predicts the next token rather than looking up an answer. How that objective enables useful work—and why fluent output is not proof.

  2. 2
    Tech Note

    What changes when an LLM has more parameters? (Part 2)

    Parameters are trained numbers, not stored facts. Model size only makes sense alongside data, compute, post-training, and inference cost.

  3. 3
    Tech Note

    Why is a long context window not memory? (Part 3)

    A context window is working space for one request. Separating parameters, chat history, memory, and RAG shows why selection matters as much as length.

  4. 4
    Tech Note

    Does longer reasoning make an answer better? (Part 4)

    Reasoning helps decompose problems; it is not proof. Separating hallucination types and task-specific checks makes model results explainable.

  5. 5
    Tech Note

    When does an LLM become an agent? (Part 5)

    An LLM proposes actions; an executor calls tools. Separating model, tools, state, and control loop reveals what reliable task completion requires.