AI: Meta Leans in with LLaMA2

...Open Source competition to heat up

It’s been anticipated for a while now, but the drumbeat continues on Meta releasing a commercial license version of its open source LLaMA 2 soon:

“With LLaMA V2, Meta may be trying to benefit from the open-source community, similar to what Google has done with Android.

The Financial Times, citing three sources familiar with the project, reports that Meta wants to launch a commercial AI model to compete with OpenAI, Microsoft, and Google. The model is said to generate language, code, and images.

It may be a new variant of Meta's LLaMA, a large language model used in numerous open-source projects. LLaMA v1 has only been released since March, under a research license and therefore may not be used directly for commercial purposes. However, replicas exist.”

It’s notable because it continues to press on Meta’s key advantage on Open Source LLM AI models vs OpenAI/Microsoft, Google and others. They don’t plan to make money off their models via API fees, or subscriptions. 

So as outlined in earlier posts, Meta is well positioned to leverage AI across its multi-billion users across Instagram, Facebook, Messenger and WhatsApp. Not to mention what they’re doing with Threads against Twitter. An open source LLaMA 2 with commercial options, is just another way for Meta to throw sand in the gears of competitors who have a different business model. 

As the piece above continues:

“According to the Financial Times and an insider report from June, the new model will remain open source. The details of the model will be published and can be used for further work.

However, Meta could reserve the right to license the model commercially and offer additional services to companies, such as fine-tuning it with its data, which the open-source community cannot easily do.

In this way, Meta would benefit twice: The open-source community would develop the model and drive AI applications based on Meta's technology. An outsourced, free research department, so to speak. The LLaMA ecosystem could grow quickly.

If Meta reserves commercial licensing, it has the sole right to make money from the model. It's a bet that companies will do business with Meta despite its open-source availability because the total package is more attractive than open-source-only use.

Whether and for which LLaMA services Meta might charge in the future, such as the aforementioned fine-tuning, is still unclear, according to Financial Times sources. The main goal of the new model is to break the dominance of OpenAI, says an insider familiar with the company's high-level strategy.”

It’s a good set of chess moves for Meta vs OpenAI/Microsoft and Google in particular. And the company is focused on these efforts from Founder/CEO Mark Zuckerberg down.

Fostering open source innovation faster in the ecosystem, feeds the concern that Google engineers highlighted a few months ago via a leaked memo, on Google/OpenAI ‘having no moat’, with open source development accelerating innovation beyond the big commercial LLM AI models to potentially counter. Google’s new global head of AI Demis Hassabis is not as concerned, but these are early days for LLM AI model innovations.

It’s going to be a busy few months. Stay tuned.