xAI just shitcanned 500 data annotators because Musk finally looked at the bills. Turns out paying humans to label cat pictures costs real money - who could have predicted that? These people were doing the grunt work that makes AI possible, telling computers whether an image shows a cat or a sandwich. Mind-numbing work, but someone has to teach these models basic pattern recognition.
Memphis Power Nightmare
That Memphis facility is burning through electricity like my old Bitcoin mining rig, except this one costs millions per month. Expensive H100 GPUs running 24/7 while they figure out why their power bills look like phone numbers. They were creating actual tech jobs in Memphis until Musk decided 500 people was too many.
Annotation Hell is Real
I've been through ML projects where annotation costs spiral out of control. You start with "we need a few thousand examples" and end up needing millions. Each labeled example costs $0.10-0.50 depending on complexity. Do the math - xAI probably burned $50M+ just on labeling before someone said "fuck this."
The plan now? Hire 5,000 "specialist AI tutors" instead. Same shit, different title, 10x more people. Fire everyone, then hire different people for the same work. Classic Musk pattern - cut costs first, figure out the consequences later.
Everyone's Winging It
Nobody has AI training figured out. OpenAI throws armies of humans at RLHF feedback loops. Anthropic does their constitutional AI thing that nobody understands. Google just feeds the entire internet into Gemini and hopes for the best.
Specialized training probably works better than random Mechanical Turk contractors, but it costs 10x more per person. Basic economics that apparently took a rocket scientist this long to figure out.
The 500 people who got fired learned the hard way: when the AI gold rush hits scaling problems, data workers get cut first. Same story as every tech boom - the people doing the actual work get axed when the spreadsheets don't add up.