Tired of being an afterthought in the AI race, MEta just took a big swing to get ahead.
What happened: Meta is granting open-source access to LLaMA 2, the large language model that powers its AI technology, making it freely available to copy, change, and be used for research and commercial purposes.
Catch-up: Meta had previously open-sourced its last model, LLaMA, for select researchers. The model was promptly leaked and spurred rapid chatbot creation.
Why it matters: Much like open relationships, open AI models come with opportunities and risks. As The Economist notes, a future where powerful AI models are open-source would:
- On one hand, reduce the monopolistic control of a handful of big tech companies, make AI more accessible, boost transparency, and accelerate AI innovations.
- On the other hand, it would make it easier for bad apples (including hostile regimes) to manipulate AI systems for treacherous purposes, while making regulation harder.
What’s next: Meta’s move could force its AI competitors—Google and OpenAI—to open-source their own models in order to retain users and the data that comes with them.
- This potential outcome was outlined in a leaked letter from a Google employee titled "We Have No Moat, And Neither Does OpenAI" detailing why open-source AI will surpass the two companies.
- “We have no secret sauce,” the letter reads, “...people will not pay for a restricted model when free, unrestricted alternatives are comparable in quality.”
Bottom line: According to Meta’s AI head Yann LeCun, AI models could end up being like the internet—the most famous thing that runs on open-source software—where “open, communal standards” form a foundation that private enterprises build atop of.—QH