OpenSourceAI

Open Source AI

⚠️ Sample. A small knowledge base mapping the debate over what "open source" should mean for AI (the example corpus shipped in this repo — kept deliberately small so the whole thing is readable end-to-end).

Overview

Open Source AI is an AI system released under terms that grant users four freedoms — to use, study, modify, and share the system for any purpose without asking permission. The OpenSourceInitiative codified this in the Open Source AI Definition (OSAID) 1.0, released in October 2024, which requires access to three components: data information, complete source code, and model parameters (weights). The definition adapts the 1998 Open Source Definition to machine-learning systems, where the software alone is insufficient to reproduce or modify a model. It is deliberately binary — a system either meets the freedoms or it does not — and is intended in part to curb OpenWashing.

Connections

• OpenSourceInitiative — author and steward of the OSAID
• OpenWeights — the partial-release middle ground the definition does not accept as fully open
• OpenWashing — the mislabeling practice the definition aims to constrain
• TrainingData — the most contested component of the definition
• ModelLicensing — the broader licensing context the definition sits within