SIGNAL//DESK
AI/MLsrc: curated AI glossary

Foundation model

A foundation model is like a 'jack-of-all-trades' AI that has been trained on a massive amount of information so it can handle many different jobs, like writing, coding, or summarizing. The 'frontier' models are simply the newest, smartest versions of these tools that lead the pack in performance.

A foundation model is a large-scale neural network pre-trained on diverse, unlabeled datasets that can be adapted to a wide range of downstream tasks through fine-tuning or in-context learning. 'Frontier models' refer to the current state-of-the-art class of these systems that exhibit the highest levels of capability and scale.

A model pre-trained on broad data at scale that serves as a base for many downstream tasks via prompting or fine-tuning; 'frontier model' denotes the most capable current generation.

evolution

  1. 2017-06 · history
    Transformer Architecture

    Google researchers introduced the Transformer, providing the scalable self-attention mechanism that enables modern large-scale pre-training.

  2. 2018-06 · history
    GPT-1 Release

    OpenAI demonstrated that a generative pre-trained transformer could be fine-tuned to achieve high performance across diverse NLP tasks.

  3. 2020-05 · history
    GPT-3 Launch

    The release of GPT-3 proved that scaling model parameters to 175 billion enabled emergent few-shot learning capabilities without task-specific fine-tuning.

  4. 2021-08 · history
    Coining the Term

    Stanford researchers formally defined the term 'Foundation Model' to describe the paradigm shift toward large, multi-purpose pre-trained systems.

  5. 2022-11 · history
    ChatGPT and Mainstream Adoption

    The release of ChatGPT popularized the use of foundation models through an accessible chat interface, establishing the 'frontier' model standard.


← all terms