Get the latest tech news

Contra Zuckerberg on 'Open Source AI'

‘Open Source AI’ Friday, July 26th, 2024 In internet tradition, I am putting aside my other duties and taking the day to write.1 Clearly Mark Zuckerberg’s letter, “Open Source AI Is the Path Forward”, accompanying the exciting release of Llama 3.1, was not intended to withstand intellectual critique. It was meant to rally goodwill and support for Meta AI from AI developers, many of whom believe in open source and software freedom.

In the theme of disanalogies to the history of Linux, Zuckerberg is proposing an ecosystem where anyone in the community can participate in a surface-level “Bazaar” and contribute tuning data sets and methods for patching shallow issues, but fundamental improvements can only come down from the “Cathedral” on high. To my knowledge, many such scenarios postulate that the risks come from the unexpected emergence (or revelation) of radical new model capabilities at some stage of training, coupled with inadequate ‘alignment’ and ‘evals’, leading to sudden loss of control of an AI system somewhere in the ecosystem, with catastrophic consequences. An alternative way of reading this letter is a scathing critique of Facebook and Instagram, these being examples of Meta products that promise benefits to society while working hard to trap them into a closed ecosystem and using closed-source AI to subtly mine their behaviour and undermine their psychology over time, fuelling an advertising machine that is the very reason “openly releasing Llama doesn’t undercut our revenue, sustainability, or ability to invest in research like it does for closed providers.” If there’s one topic on which Zuckerberg has no credibility, it’s whether some technology is a net good for society.

Get the Android app