Meta is planning to release a commercial version of its AI model which will allow companies to build their own software using Meta’s AI technology.
Meta has been coming a distinct third in the AI tech race behind rivals OpenAI and Google and this move signals its intent to catch up. Earlier in 2023 Meta gave researchers and academics access to its language model, LLaMa. Meta’s website still says you need to apply for access, ironically by filling out a Google Doc.
A week after Meta started accepting applications from researchers to use LLaMa, someone posted a link on 4chan to a downloadable copy of the entire package. So the Llama actually bolted some time ago. The imminent official release will now give wider, legitimate access to companies and startups that are looking to incorporate generative AI into their software.
Meta’s decision to keep its AI open-source is a key differentiator between its offering and that of Google and OpenAI, both of which have kept their models proprietary.
The open-source nature of Meta’s offering will likely drive faster adoption albeit with its prospects for monetization being unclear. The other benefit of going the open-source route is that it allows companies to improve, debug, and customize the software themselves.
Meta has been criticized recently for how it has handled privacy and misinformation issues so going the open-source route will probably help with public perception.
What will access to LLaMa cost?
Nick Clegg, Meta’s global affairs chief, said that “openness is the best antidote to the fears surrounding AI”. With persistent skepticism surrounding the ethics and dangers of AI that’s probably a fair point.
What isn’t clear is how Meta plans to make money from its model. Meta shareholders are probably wondering about that too, especially as they’re still bleeding money from Zuckerberg’s Metaverse project.
For now, access is to remain free with talk of possibly charging companies if they want to fine-tune the models with their own data. With Meta offering free access to its model, it may look for opportunities to sell services on top of the model rather than control access to it.
Google and OpenAI aren’t happy with the open-source offering from Meta. They’ve raised the risk of potential bad actors customizing the AI for nefarious ends or circumventing ethical filters.
Their main concern though is probably whether people will continue to pay to use their models when there’s a really good free option available.