Finding a harder problem
Most financial advisors measure success by the quality of the advice they give.
Marc measures it by something else entirely: whether he can make an AI model fail.
"I'm proud every time I'm able to cause the model to fail," he says. "It's quite the challenge."
It's an unexpected use for a Series 65 license. Instead of advising clients, he's helping train frontier AI systems by designing financial scenarios that expose the limits of their reasoning. Every failure reveals a blind spot. Every blind spot becomes an opportunity to improve the model.
That's what keeps him coming back.
"I can see first-hand how the models are improving day by day, sometimes even by the hour, because of the work we do."
Creativity over correctness
The job turns out to require far more imagination than most people expect.
Instead of checking answers all day, he spends much of his time inventing finance scenarios that force models into unfamiliar territory.
"I enjoy creating high-level finance scenarios and testing the models through them," he says.
The hardest part isn't making a task difficult. It's making it novel.Every scenario is designed to answer a simple question: What hasn't this model learned yet?
"I often make the judgment that a particular task I'm thinking of is not creative enough or not challenging enough. I have to switch gears and think creatively and intricately to provide the best training."
Drawing from years of experience
Good prompts, he says, are built from experience.
Years in advisory and consulting exposed him to a steady stream of businesses, clients, and financial decisions. Today, those experiences resurface in unexpected ways; not as case studies, but as the foundation for entirely new scenarios designed to push AI beyond familiar patterns.
"My experience has enabled me to write numerous tasks and develop many creative financial scenarios that truly test the model," he says.
It's one reason the work resists becoming formulaic. Every new client, industry, or business problem becomes another way to probe the edges of what a model understands.
From skepticism to optimism
Like many people, he didn't come into the work convinced AI would make professionals better.
If anything, he worried the opposite might happen.
"I was skeptical," he says. "I felt that leaning too heavily on AI could be counterproductive to your own critical reasoning skills."
Working with frontier models every day gradually changed that view. What he found wasn't a replacement for critical thinking, but another tool that demanded it.
The professionals getting the most out of AI, he believes, aren't the ones outsourcing their judgment. They're the ones learning how to challenge the models, build on their strengths, and know when to trust their own instincts instead.
Still waiting for AI to know when not to think
For all the progress he's watched firsthand, one challenge still feels unsolved.
Sometimes, he says, the smartest models think too hard.
Ask an advanced reasoning model a simple question, and it may spend far longer than necessary arriving at an answer. In those moments, a quick web search is still faster.
His ideal system wouldn't just reason more effectively. It would know when deep reasoning isn't needed at all—switching seamlessly to a faster model for simple questions, then returning to more sophisticated reasoning when the problem calls for it.
"Maybe we can get a project for that," he jokes.
Advice for new experts
Ask him what advice he'd give someone joining Mercor, and he doesn't talk about AI.
He talks about the guidelines.
"My advice would be to learn the project guidelines like the back of your hand," he says. "They're dense and they're complex."
Then comes his most practical recommendation:
"Control + F is really, really useful."