I thought about titling this “What have I been up to this year,” but who can resist a bit of clickbait?
Since January 2025, I’ve been working with Mercor (pardon the shameless referral link) as a human data trainer and reviewer:
As a trainer, I use my subject matter expertise in software engineering to help AI get better.
As a reviewer, I use my empathy and my obsessiveness over fine details to help other trainers get better at making AI get better.
I wrote a Medium post about working in human data if you’re curious about what’s involved, but it’s a trending topic and I wanted to give my followers a more personal vibe for what it’s like.
The whole thing started as a side gig that I didn’t anticipate doing for long, a way of building up cash after I ran out working on my last startup. But now I’ve been wondering whether this might be a third (or fourth) career for me.
The most visible AI jobs are positions out of reach for mere mortals: the researchers in San Francisco and the AI engineers who are implementing the hard tech. I think that “human data trainer/reviewer” might be one of the first AI-native jobs that’s accessible to the rest of us all around the world.
I work on projects that involve large groups of people from around the world, all of us contributing our expertise. I don’t know of many gigs where you regularly interact with smart people from every industry all at once. Some projects are more narrowly tailored, in which case I’m more closely working with software engineers, but still with people from everywhere. Everyone’s smart and the work is super engaging.
The pay so far has been phenomenal. I’ve posted screenshots of my payouts on LinkedIn and people still don’t believe me.
Not everyone is being paid that much, sure, but it’s still top dollar for your experience level and industry. The flexibility is unmatched. I have dozens of people asking me every week on LinkedIn how to apply and whether it’s real. Many people think it’s a scam. (It’s not.)
What makes a career? I think there has to be an abundance of work, and it certainly doesn’t appear like Mercor or other human data companies are going out of business any time soon.
There also needs to be depth and progression, and here’s where it gets interesting. The more projects that I do with Mercor, the better I get at being a trainer and a reviewer. There’s a skill set involved that you hone over time, and folks gradually develop different work philosophies.
For example, I think that the cornerstone of my approach as a reviewer is empathy. I spend a lot of time trying to understand where the trainers whom I work with are coming from and their backgrounds in order to improve their contributions. I encounter so many people who are brand new to the world of machine learning and struggle with translating their considerable expertise and experience to the task of improving AI. I often play the role of an AI sherpa, bridging the gap. There’s a ton of soft skills involved.
Often I feel like I’m being paid to be an AI philosopher. I can’t go into details, but doing this job well on a daily basis means thinking very deeply about what AI is and what purpose it serves. It feels like I’m building a house but I don’t get to see what I’m building, I only get to see the final result along with everyone else. I have to have a strong sense of what’s the end goal to get there.
I did a cheesy little promo video that makes me look like a bit of a corporate shill or a wide-eyed naive AI booster, which is fine. (I’m neither of those things.)
I did it because I wanted to meet some of the people I’ve worked with in person. It’s weird, you’d think that working with a huge mass of fellow independent contractors everyday would be a deeply estranging thing, yet I’ve felt some of the most intense camaraderie doing human data work.
Probably the most frustrating aspect of what I do is that I’m sitting at the epicenter of so many important developments in AI, yet my NDA means that I can’t comment on very much.
One of those developments is the rise of human data providers themselves. Mercor is probably the most well-known and is certainly the highest-quality of the bunch, but there are others. I have an ambition to write more about this emerging field, but I’m not sure what to write or how to write about it.
That doesn’t mean that I won’t try. In the coming weeks, I’d like to write more about human data and AI. Human data as a niche inside of AI fascinates me because I believe that it points to the future of knowledge work, which will involve a lot of training and evaluating AI to make it do our bidding and eliminate repetitive work. I predict that we’ll all be doing aspects of human data work in our jobs before long.