Open source AI makes a model’s code and parameter values publicly available, often allowing users to inspect, modify, customize, download, and run it on their own infrastructure. section 3.1
A practical example
Llama models illustrate the model: users can download them, run them locally or in their own cloud environment, and customize them instead of relying only on a closed provider’s remote service. section 3.1
Choose open source AI when control matters more than a frictionless start:
- Self-host models to improve governance, inspect code, and keep sensitive data from leaving the organization. section 3.2
- Customize models more freely, including through fine-tuning. section 3.2
- Expect infrastructure and maintenance costs, and check models carefully because malicious releases can execute unintended code. section 3.2
- Treat “open” cautiously: providers may withhold training data, training logic, or other assets needed to reproduce the model exactly. section 3.1
Table 3.1 Proprietary vs. open source AI
| Proprietary AIBest for . . . |
Open source AIBest for . . . |
|---|---|
| • Done-for-you solution • Easy start • No specialized knowledge required • Small-scale use, in which pay-as-you-go AI is cheaper than maintaining your own infrastructure |
• Self-hosting so that you enjoy better governance and transparency • Large-scale use, in which maintaining your own infrastructure is cheaper than pay-as-you-go AI • Model customization (e.g., fine-tuning) |
Compare open source and proprietary AI by deployment control, scale economics, ease of use, and customization. section 3.2
