In the AI Era, the Definition and Strategic Evolution of Open Source... Not Necessarily "Free"

As Meta distributes its AI model Llama as open source while applying revenue-sharing licenses to specific companies, discussion about the concept of open source and commercial utilization is heating up again. Many companies and users have the perception that "open source" means "freely and freely usable," but in reality there are many cases where costs occur or commercial restrictions exist depending on license terms. The open source movement origin: Richard Stallman founded the Free Software Foundation in 1985 establishing four fundamental freedoms (freedom to use, study, modify, and distribute); Linus Torvalds released Linux kernel as open source in 1991 demonstrating that open development produces high-quality software; the Open Source Initiative (OSI) formalized the open source definition in 1998. Open source license spectrum: (1) Permissive licenses (MIT, Apache 2.0, BSD) -- maximum freedom; anyone can use commercially, modify, and redistribute without requiring changes to remain open; (2) Copyleft licenses (GPL, LGPL) -- modifications must be released under the same license; commercial use permitted but derived works cannot be proprietary; (3) Business Source License (BSL/BUSL) -- time-delayed open source; code becomes fully open after a period (typically 4 years) but commercial use restricted in the interim; (4) Custom AI licenses (Meta Llama Community License) -- open weights but commercial restrictions above certain revenue thresholds; not OSI-approved "open source" by strict definition. The Meta Llama case: Llama 2 and 3 are "open weight" models (the trained model parameters are publicly available) but not fully open source (the training code and data are not released; commercial use restricted for companies with 700M+ monthly active users); companies like Microsoft and Amazon using Llama in cloud services must negotiate separate commercial agreements with Meta. The strategic value of "open source" AI: ecosystem development (developers building on your model create network effects); talent attraction (researchers want to work on models they can publish about); competitive pressure on proprietary competitors (OpenAI); data collection (usage patterns from open deployments inform model improvement). The misconception cost: organizations assuming Llama or similar "open" AI is completely free may face compliance issues if they exceed license thresholds or use the model in ways the licensor prohibits.