Perfilado de sección

  • LESSON 5 - ARTIFICIAL INTELLIGENCE FOUNDATIONS FOR BROADENING CULTURAL HORIZONS

    How Do Large Language Models Learn?

    In this lesson, we delve into the inner workings of language models. How do they learn the meaning of words? Through massive volumes of data and the constant prediction of the next word in a sequence, LLMs recognize patterns and semantic proximity.However, generating plausible sentences is not the same as understanding or guaranteeing their truth. What does it mean for a response to be statistically probable, but not necessarily correct?

    We also analyze the risks: implicit assumptions, reproduced stereotypes, and confusion between entities. Finally, we distinguish between a base model (trained to complete text) and a model adjusted via fine-tuning, aligned to follow specific instructions. This lesson invites us to understand that behind every answer lie training, classification, and alignment processes, and that knowing these mechanisms is essential for using these tools with pedagogical judgment.

    At the end of the lesson, you will be asked to answer a few questions. 

    👨‍🏫 Instructor: Guido Ivetta

    • How Do Large Language Models Learn? URL

      Prefer to watch in a language other than Spanish? Just turn on CC, go to Settings (⚙️) > Subtitles > Auto-translate, and select your language!

    • Below you will find the lesson slides, designed as a reference to revisit the ideas, questions, and tensions raised in the video. (Please note: The slides are in Spanish).

      📚 Key Concepts

      Alignment: The final stage where humans evaluate and rank the model's responses. It serves to help the AI adopt human values, more natural ways of speaking, and avoid undesired behaviors.

      Instruction Tuning: The process of training the model with "Instruction-Response" pairs. This is what allows the model to understand that if you ask it something, it should provide an answer rather than continue asking questions.

      Semantic Proximity (Embeddings): The way the model organizes words in a mathematical space. Words with similar meanings (such as "chat" and "talk") end up "close" to each other.

      Probability Distribution: When faced with a blank space, the model does not make a random guess; instead, it assigns a percentage probability to each word in its vocabulary (e.g., "dog" has a 90% probability compared to "barking").

      Base Model: The initial version of the model after processing a massive amount of data. Its primary function is to predict the next word, but it does not yet know how to "obey" commands; it only completes text.

      A creation by Fundación Vía Libre in collaboration with FAMAF – UNC.
    • Assignment: How Do Large Language Models Learn? Cuestionario