Uber hat Doktoranden abgeworben, um sich dem „Project Sandbox“ anzuschließen. Nach einem Monat hieß es, dass ihre KI-Trainingsverträge abgelaufen seien.
https://www.businessinsider.com/uber-project-sandbox-shows-ai-training-contractors-the-door-2025-11
7 Kommentare
Twelve contractors on a project were let go two months early. So? Why is this news?
Typical project management: they had a year long project and a budget for one PHD to do it. As any good PM would, they hired 12 PHDs and declared the project done in 1 month.
Ahahahaha! As a PM myself I approve this.
Most PHDs aren’t in high demand because their advanced degree work, unless it was done in a corporate sponsored lab, is very different from real world work
And even this example kind of shows that. Google is paying Uber for AI training. Why does Google need help when they have the brightest minds in deep mind and elsewhere
This was probably labeling or some other mind numbing job
This is what „AI strategy“ looks like for a lot of big firms right now: treat researchers like options, not people. You buy a bunch of PhDs for a few months, see if anything magically pops, then mark the whole thing to marketing and move on. The external narrative is „bold long term bet on AI,“ the internal reality is „cheap optionality with zero commitment.“
The messed up part is the asymmetry. Uber gets the IP, the signal to investors, the press line about cutting edge sandbox projects. The researchers get a weird gap on their CV, an NDA, and maybe a non-compete if they’re unlucky. It’s basically labor arbitrage wrapped in innovation theater.
If companies were serious, you’d see multi‑year research tracks, clear publication policies, and predictable conversion paths, not one month „training contracts“ that vanish overnight. Everyone keeps talking about AI risk at the model level, but this kind of behavior is exactly the social risk: incentives to burn through talent faster than you can build any real institutional memory.
> Tasks ranged from annotating photos and videos to evaluating whether an AI-generated answer to a prompt was accurate and complete, the workers said.
> Pay varied. In one case, base pay was $55 an hour, rising to as much as $110 an hour, or around $19,000 a monthccurate and complete, the workers said.
Um… (blinks) wut?
TLDR. One dozen contractors got let go 2 months early from their contract. Wooopeeedo. I guess the real story was either that they were consulting for Google OR that their AI division is closing.
Isn’t this Mercor’s whole business model? You get hired on a temp basis to annotate llm output based on your background and get paid by the hour.