Open source and vendors
The AI ecosystem mixes **downloadable weights**, **hosted APIs**, and **proprietary stacks**. “Open” is a spectrum: OSI’s Open Source AI Definition asks for freedoms to use, study, modify, and share—with preferred forms for modification ...
What it is
The AI ecosystem mixes downloadable weights, hosted APIs, and proprietary stacks. “Open” is a spectrum: OSI’s Open Source AI Definition asks for freedoms to use, study, modify, and share—with preferred forms for modification including data information, code, and parameters. Many “open models” are open weights under custom licenses, not OSI open source.
<!-- IMAGE: spectrum: proprietary API ←→ open weights ←→ OSI open source AI -->
Visual Spec & Architecture Diagram
Open spectrum continuum: Closed API-only → Open weights (restricted license) → Open weights + code → OSI-open AI definition aligned. Tick marks with examples as generic categories not endorsements.
Why it matters
Licenses, cost, privacy, customization, and compliance differ. ANN Learn stays vendor-honest: teach tradeoffs; don’t pretend one logo wins forever. Dates matter—terms change; always re-read the primary document before shipping.
How it works (plain)
Compare options on: capability, price, latency, data retention, region, fine-tune options, audit needs, and exit plan. Read three layers:
- Weight/code license (e.g., Llama Community, Apache 2.0, RAIL).
- API / service terms (OpenAI, Gemini, AWS, Microsoft, Anthropic privacy/commercial docs).
- Acceptable use / prohibited use policies.
This course is legal literacy, not legal advice. Product decisions need counsel.
Everyday example
Cooking with your own kitchen vs ordering delivery—control vs convenience, different bills and different rules about what you may resell.
Try it
For one use case, score API vs local open weights on privacy, cost, and maintenance (1–5 each). Note which primary URLs you would re-check this quarter.
Myths
- ⚠️ Myth: Open weights = OSI open source.
- ✓ Reality: OSAID requires more than downloadable parameters (data information + code + freedoms).
- ⚠️ Myth: Open source always means safer.
- ✓ Reality: Safety depends on deployment, policies, and ops.
- ⚠️ Myth: A vendor API removes all responsibility.
- ✓ Reality: Your tools, prompts, and user-data handling still matter.
Sources
- OSI Open Source AI Definition: https://opensource.org/ai/open-source-ai-definition ↗
- Hugging Face licenses on Hub: https://huggingface.co/docs/hub/main/en/repositories-licenses ↗
- SPDX license list: https://spdx.org/licenses/ ↗
- OpenAI Service Terms: https://openai.com/policies/service-terms/ ↗
- Gemini API Additional Terms: https://ai.google.dev/gemini-api/terms ↗
- Anthropic Privacy Center: https://privacy.anthropic.com/en/ ↗
- AWS Service Terms: https://aws.amazon.com/service-terms/ ↗
- Microsoft licensing terms (EAEAS changes): https://www.microsoft.com/licensing/terms/en-US/product/changes/EAEAS ↗
