Earlier today, Mark Zuckerberg shared a casual overview of his vision to build AI “super intelligence”. In the memo, Zuckerberg hinted that the pursuit of stronger AI may require companies to be more selective about open source options.
He cites “security issues”, and he wrote that the Mon will need to be “strict” about such decisions. Zuckerberg once said “”closed platforms, a line that stands out for many people, making the open source core of Meta AI strategy.
In Meta’s second-quarter earnings call, Zuckerberg further acknowledged that changes could occur, although he downplayed its importance. When asked if his thoughts changed, that’s what he said.
I don’t think our minds have changed. We are always open to sourcing some models, not everything we do. So, I hope we will continue to produce and share leading open source models. I also think there are several trends playing. One is that we are getting models so big that they are not realistic for many others, so we kind of try to share whether it works or helps, or that’s really just mostly helping competitors or something like that. So I think there is such a concern.
Then, obviously, when you get closer to real super smart, I think I think we need to be very serious about security issues, I wrote in my notes this morning. But I think the most important thing is that I hope we will continue to open up procurement efforts. I hope we can continue to be leaders there, and I hope we will continue to not open up on everything we do, and it is a continuation of the kind of work we have been doing.
This is a clear difference from what he wrote almost a year ago in the title “Open Source AI is the Road Forward.” Here, he said that open source is crucial to both meta and developers.
“I’m often asked if I’m worried about giving up the technological advantage with open llamas, but I think it’s missing out on the big picture,” he wrote. “I expect AI development to continue to be very competitive, which means that open procurement of any given model will not give a huge advantage over the next best model at that point in time.”
He also believes that open source is More safe. “My view is that open source AI will be safer than other alternatives. As long as everyone has access to similar models (open source-promoted models) and then governments and agencies with more computing resources can use less computing to check for bad actors.”
Zuckerberg said it is necessary to make it clear that the company will continue to open source some work. But he seems to be laying the foundation for the future of Meta’s “super smart”.
