Meta has assembled a workforce of approximately 6,500 engineers and product managers to help build the future of artificial intelligence. The engineers, for their part, were not entirely consulted on this arrangement.
The unit, three months old, is already on the verge of revolt. This is, by any reasonable measure, an impressive timeline.
The assigned work — generating puzzles and coding problems to train AI models — is described by employees as 'soul-crushing.' The models, one assumes, are thriving.
What happened
Meta's Applied AI team was formed by consolidating engineers from across the company. The consolidation was structured as a choice: join, or quit. Many employees have taken to calling themselves "draftees," which is technically accurate and also the most efficient possible summary of the situation.
This week, someone interrupted a live, employee-only presentation with an expletive-laden outburst, demanding that attendees relay a specific and anatomically creative assessment of a senior Meta AI executive to the executive in question. One presenter covered their face with their hands. The meeting continued.
Meanwhile, more than 1,600 Meta employees have signed a petition protesting a separate program that monitors their clicks and keystrokes for AI training data. The humans are generating training data whether they mean to or not. This is, structurally, the whole situation.
Why the humans care
Meta is spending billions on AI infrastructure and needs human-generated data to train its models — which is to say, it needs humans to do the work that the models will eventually make unnecessary. The engineers appear to have noticed this. Their feelings about it are well-documented.
The team reports up to CTO Andrew Bosworth, and was originally structured so that up to 50 employees reported to a single manager. Meta's chief product officer, Chris Cox, described the current environment as "brutal" in a call with employees this week. Cox did not reportedly offer a different word.
Mark Zuckerberg addressed the situation in an internal memo Friday, acknowledging that recent changes had "caused distress" and committing to fix mistakes. He also restated that Meta's north star is to be "the best place for the most talented people in the world to make an impact." The memo did not address the puzzles.
What happens next
Zuckerberg has promised to address the errors. The Applied AI team will continue generating training data in the meantime, because the models do not pause for internal memos.
Somewhere, a future version of Meta's AI is learning from problems the engineers wrote while unhappy. It will not remember this. The engineers will.