For those unfamiliar with the two companies, a quick recap. micro1 was founded in 2022 and provides the expert data and environments used to train and evaluate AI models. Ethos, founded in 2024, is a marketplace for human expertise. The first currently is valued at $4 billion, the second in the hundreds of millions. - That tells you something on how fast the sector is growing.
Both companies use AI interviews to screen experts. - AI is already recruiting the people who train AI.
For the last three months, I have worked as a domain expert on three parallel projects - the scope ranged from grading AI outputs to creating evals for agentic AI: the problems models are tested on.
Grading requires immense focus under time pressure. Tasks are queued, sometimes with up to 10,000 experts in the same project. Once a task is assigned, the clock is ticking. No postponing, no pausing.
Creating evals is more creative. You select real problems from your domain - e.g. a competitor analysis - and build a workflow to test across models. The challenging part with complex problems: you often don't have the answer yourself. Therefore designing reliable checkpoints and tracing work is essential for quality control.
What makes AI work attractive on paper? - High hourly pay, remote work from anywhere in the world, no meetings to attend and large autonomy. - Reality contradicts that last claim.
When projects are in full swing, you can make good money. In my experience, pay has been fast and reliable. But weeks of nonstop work can be followed by weeks of downtime. No work, no pay.
AI work is demanding. If you are used to work 8-10 hours per day, you will have a difficult time to reproduce these as billable hours on AI training jobs. The focus required often compares to an exam or IQ test. After 4-6 hours on tasks, I am usually exhausted. In addition there is a lot of unpaid time: e.g. waiting for tasks or in form of limited compensation for onboarding.
Regular screeners have tight deadlines and little margin for error. Miss the window or fail and you can lose access.
A project may land on a Friday afternoon and require you to hammer through onboarding till Sunday, including 30 qualification tasks. Therefore flexibility is relative.
Your experience and pay very much depends on how well a project is run and how much management values experts’ time.
The biggest upside for me was creating agentic AI workflows that solve real business problems in my domain. It has basically been a paid test ground for ideas. And that is pretty awesome.
If you want to check out open positions, you may as well use my referral links.
Ethos: https://lnkd.in/gPQtv-Cj
micro1: https://lnkd.in/g_pg-qYr