Technology

Foundational Models

Massive, pre-trained AI models: they serve as the adaptable base (foundation) for numerous downstream applications via transfer learning.

Foundation Models (FMs) represent a paradigm shift: a single, massive AI model trained on a vast, general dataset. This pre-training allows FMs to be adapted (via fine-tuning or prompting) for a wide array of specialized tasks: natural language processing (NLP), image generation, and code completion. Key examples include OpenAI's GPT-4, a transformer architecture with an estimated 1.7 trillion parameters, and Google's BERT. The core value is transfer learning: FMs eliminate the need to train task-specific models from scratch, significantly cutting development time and compute resources for new applications.

https://crfm.stanford.edu/

What builders pair with Foundational Models

Projects using both technologies. Select a pairing to see a project.

Built with Foundational Models

AI personas 1 shared project Coating 1 shared project Data curation 1 shared project Embeddings 1 shared project function calling 1 shared project GPUs 1 shared project

6 more pairings

Open source models 1 shared project Patterning 1 shared project Prompt Engineering 1 shared project Semiconductors 1 shared project Supervised learning 1 shared project Zero-shot learning 1 shared project

Pairing: AI personas

AI agents for investment research

Los Angeles · May 21, 2024

view project | view AI personas

Recent Talks & Demos

Showing 1-3 of 3

Members-Only

Sign in to see who built these projects

Sign in View FAQ

NParks Zero-shot Wildlife Recognition

Singapore Jan 10

Foundational Models Embeddings

Hacker Fab: DIY Semiconductors

Semiconductors Patterning

Prompt Engineering Investment Agents

Los Angeles May 21

function calling Prompt Engineering