In Llama 2's commercial terms, Meta says companies with 700M+ MAUs must request a license, and users are prohibited from utilizing Llama 2 to improve other LLMs

Although Meta has announced that it will open-source its LLM and allow free usage for commercial and research purposes …

Moneycontrol 2023-07-20 Arun Padmanabhan

Context & Ripple Effects

Meta positioned Llama 2 as available for research and commercial use in its initial Llama 2 release, but its community license defined where that availability stopped. The terms make the model open-weight rather than unrestricted.

The boundary mattered as businesses began building on the model: Alibaba's planned Llama 2-based applications showed that commercial adoption could extend beyond Meta's own products while remaining subject to Meta's conditions.

First-order effects

Companies with at least 700M monthly active users must seek a separate license from Meta before using Llama 2, creating an approval step for the largest platforms.
Llama 2 users cannot use the model to improve another LLM, limiting reuse by developers whose work could feed competing foundation models.

Second-order effects

Enterprise adopters and distributors must assess their user scale and intended model-development workflows before committing to Llama 2, rather than treating its commercial availability as unconditional.
Rival model providers can differentiate on whether their licenses permit model-to-model development, while Meta retains leverage over the largest potential downstream users.

Third-order effects

If such terms become standard, open-weight AI will be commercialized through layered licenses: broad access for application building, with tighter control over strategic reuse and the biggest distributors.
The practical divide in AI access may increasingly be set by license compliance and scale thresholds, not simply by whether model weights are publicly available.

The trend: This is one example of open-weight AI shifting toward controlled commercialization, where wide distribution coexists with restrictions on competitive model development and platform-scale use.

Discussion

@kyliebytes Kylie Robison on x
if you're plugged into the open source community, you likely saw everyone reading the fine print on meta's LLaMa 2 announcement the fine print says that firms over 700MM monthly active users have to obtain a license eg: snap hit 750 million MAU this year https://fortune.com/...
@hayskali Kali Hays on x
Meta gone from “Llama 2 is open source!” to telling us ok so it's “a bespoke commercial license that balances open access to the models” That does not hit the basic def of open source, director of the Open Source Initiative said. w @rosaliechan17 https://www.businessinsider.com/ …
@samandhawk @samandhawk on x
@QuinnyPig As with many things in this license agreement, it is very vague. If you use Llama 2 to *verify* or measure another LLM, I would have to imagine this is considered “improving” the LLM. Also, my pet peeve is how they are not explicit about how they measure 700M+ MAUs.
@quinnypig Corey Quinn on x
“You can't use Llama 2 to train other LLMs.” “Okay Facebook, what did you use to train Llama 2?” *crickets*
@adymitruk Adam Dymitruk on x
Llama 2 has a clause where you need a license if you have more than 700 million users which destroys Amazon, Apple, Microsoft etc from using it. I love it!
@mikemaccana Mike MacCana on x
@bgurley ... Llama 2 is not open source. The license is at https://ai.meta.com/... and it violates the open source definition. [image]

Chronicles