WEBVTT - The $10 Billion Startup Training AI to Replace the White-Collar Workforce 0:00:00.800 --> 0:00:03.920 The ten billion dollars start up training AI to replace 0:00:04.000 --> 0:00:08.080 the white collar workforce. Mercy is promising to replicate most 0:00:08.160 --> 0:00:12.120 professional work. It was also co founded by twenty somethings 0:00:12.160 --> 0:00:16.720 who previously never held a real job. By Tom Foster 0:00:17.280 --> 0:00:22.160 read aloud by Mark Lydorf. When Tasha Kozak, a social 0:00:22.160 --> 0:00:25.560 worker for the Hillsborough County Public Schools in Tampa, met 0:00:25.600 --> 0:00:28.600 the family, they were living in a car. The three 0:00:28.680 --> 0:00:32.640 children's grades had been slipping, The mother was exhausted, the 0:00:32.680 --> 0:00:36.159 father was out of the picture. Kozak began helping them 0:00:36.200 --> 0:00:39.440 connect to housing resources. She checked in with the children 0:00:39.479 --> 0:00:42.320 every couple of days at school. She called the mother 0:00:42.440 --> 0:00:45.840 every three or four. After several months, the family found 0:00:45.880 --> 0:00:50.000 stable housing. The children began to improve in school. I 0:00:50.080 --> 0:00:52.560 saw the mom got her glow back, and she started 0:00:52.560 --> 0:00:56.480 getting more consistent work shifts. Kozak says, moments like that 0:00:56.600 --> 0:00:59.520 are why Kozak does the work. She didn't set out 0:00:59.520 --> 0:01:02.320 to become a social worker, but after taking an elective 0:01:02.360 --> 0:01:04.880 in it in college, she changed her major, and more 0:01:04.920 --> 0:01:07.920 than a decade later, she still calls social work her passion. 0:01:08.880 --> 0:01:12.560 The job, she says, is about listening, connecting, helping in 0:01:12.600 --> 0:01:16.080 a way that feels instantly meaningful. In her view, about 0:01:16.120 --> 0:01:20.039 seventy percent of effective social work is that human relational component. 0:01:20.600 --> 0:01:24.960 The rest is administrative. On many evenings, after finishing her 0:01:25.000 --> 0:01:27.959 full time job with the district, Kozak logs into a 0:01:27.959 --> 0:01:31.959 website called Mercore. The San Francisco startup mercore dot io 0:01:32.400 --> 0:01:37.600 recruits workers in high skill fields doctors, lawyers, investment bankers, journalists, 0:01:37.720 --> 0:01:41.280 social workers, you name it, to help teach artificial intelligence 0:01:41.319 --> 0:01:44.679 systems how to do their work. The company places these 0:01:44.720 --> 0:01:48.160 experts in part time or temporary contract roles to work 0:01:48.200 --> 0:01:51.440 on training projects for major tech companies and AI labs. 0:01:51.920 --> 0:01:56.000 Clients that have included OpenAI, Anthropic, and Meta. It's like 0:01:56.040 --> 0:01:59.640 the Uber of advanced AI training, a gigwork platform for 0:01:59.680 --> 0:02:02.680 white collar and skilled professionals that offers a path for 0:02:02.720 --> 0:02:05.680 them to earn something extra from their expertise at the 0:02:05.760 --> 0:02:10.160 risk of eventually sacrificing their careers to AI. Like many 0:02:10.240 --> 0:02:13.880 of Mercer's contractors, Kozak first heard about the company through 0:02:13.919 --> 0:02:17.080 a LinkedIn job posting. School had just let out for 0:02:17.080 --> 0:02:20.040 the summer in twenty twenty five, so she had spare time, 0:02:20.360 --> 0:02:23.200 and she was curious about the rise of AI. She 0:02:23.280 --> 0:02:26.320 landed an interview and found herself on camera talking to 0:02:26.360 --> 0:02:29.240 an AI agent with a gentle female voice and a 0:02:29.280 --> 0:02:34.400 surprisingly natural conversational style. When Kozak described a specific case 0:02:34.440 --> 0:02:38.359 she'd handled, the agent followed up with targeted prompts, tell 0:02:38.400 --> 0:02:41.440 me more about the parent in that situation, with the 0:02:41.520 --> 0:02:44.079 kind of probing detail she was used to hearing from 0:02:44.160 --> 0:02:47.800 human supervisors. Two weeks later, she was working on her 0:02:47.840 --> 0:02:51.280 first assignments, and today she earns as much money training 0:02:51.320 --> 0:02:53.840 AI for twenty hours a week as she does in 0:02:53.960 --> 0:02:58.399 forty hours helping actual families. Social work is a very 0:02:58.520 --> 0:03:03.360 underpaid occupation, she says. Her work for Mercore is methodical 0:03:03.440 --> 0:03:06.600 and narrow. Kozak is part of a virtual team that 0:03:06.639 --> 0:03:10.040 writes prompts for an AI model to perform specific social 0:03:10.080 --> 0:03:14.639 work tasks based on fictitious case files, For instance, asking 0:03:14.680 --> 0:03:17.440 the AI to produce a social developmental history of an 0:03:17.480 --> 0:03:22.000 elementary school student who needs an individualized education plan. The 0:03:22.040 --> 0:03:25.919 case file includes notes from a parent interview, a student interview, 0:03:26.240 --> 0:03:30.000 a review of student records, and a doctor's report. Another 0:03:30.080 --> 0:03:34.000 team reviews the AI's responses, and still another handles additional 0:03:34.040 --> 0:03:37.520 pieces of the training process. There are hundreds of people 0:03:37.560 --> 0:03:41.280 on the project, Kozak says maybe more. Her team alone 0:03:41.360 --> 0:03:46.000 includes about forty contractors. Hour by hour, highly segmented, task 0:03:46.080 --> 0:03:51.040 by task, they translate professional judgment into training data. Across 0:03:51.080 --> 0:03:55.080 the US, other professionals are doing the same. In San Francisco, 0:03:55.200 --> 0:03:59.080 a doctor of internal medicine named Milania Punacha works fifty 0:03:59.120 --> 0:04:01.680 to sixty hours AGAY week as a full time pediatric 0:04:01.800 --> 0:04:04.960 hospitalist on the night shift and logs into mercor on 0:04:05.000 --> 0:04:07.520 her days off for an additional ten hours or so, 0:04:08.040 --> 0:04:11.440 asking an AI model to interpret lab results and evaluating 0:04:11.480 --> 0:04:15.480 its work. In Baton Rouge, Louisiana, the novelist and screenwriter 0:04:15.640 --> 0:04:19.840 Robin palmer Blanche started evaluating AI generated creative writing for 0:04:20.040 --> 0:04:23.919 Voice and Structure last August. Palmer Blanche, a mother of 0:04:23.960 --> 0:04:26.520 two who struggled to make a living from writing since 0:04:26.520 --> 0:04:29.480 the twenty twenty three Hollywood Writers Strike, is part of 0:04:29.520 --> 0:04:32.680 a large swath of Mercore workers who are under employed 0:04:32.680 --> 0:04:36.040 in their professions and use the platform to patch together income. 0:04:36.640 --> 0:04:40.080 She sometimes finds herself chatting with a Mercore teammate on Slack, 0:04:40.440 --> 0:04:43.920 only to realize there are other novelists doing it too. 0:04:44.080 --> 0:04:46.599 Oh my god, I read that woman's book last year 0:04:46.640 --> 0:04:49.919 and fell in love with it. Mercor has tens of 0:04:49.960 --> 0:04:53.799 thousands of such experts working on its platform, screened using 0:04:53.839 --> 0:04:57.320 those AI run interviews and project specific skills tests and 0:04:57.360 --> 0:05:01.320 practice tasks. In the three years since the company started, 0:05:01.560 --> 0:05:04.839 it's raised almost five hundred million dollars in venture capital 0:05:05.080 --> 0:05:09.640 from a who's who across Silicon Valley in finance, including Benchmark, 0:05:09.880 --> 0:05:14.919 General Catalyst, Peter Teel, Jack Dorsey, and Larry Summers. The 0:05:15.000 --> 0:05:18.440 most recent round of funding in October valued Mercor at 0:05:18.480 --> 0:05:21.320 ten billion dollars, five times what it was thought to 0:05:21.360 --> 0:05:24.599 be worth just a few months earlier. According to the company, 0:05:24.839 --> 0:05:28.040 it's been profitable since its inception. It pays out more 0:05:28.080 --> 0:05:31.240 than two million dollars per day to contractors, and it 0:05:31.279 --> 0:05:35.279 has about three hundred full time employees, largely engineers and 0:05:35.320 --> 0:05:40.679 project managers. The founders, Brendan Foodi, adarsh Hiramat and Suri Amida, 0:05:41.040 --> 0:05:44.520 three early twenties college dropouts who were buddies in high school, 0:05:44.800 --> 0:05:48.080 have become the youngest ever self made billionaires, at least 0:05:48.120 --> 0:05:53.040 on paper. Mercor's rapid rise has also brought controversy, including 0:05:53.080 --> 0:05:56.880 several class action lawsuits now moving through California courts and 0:05:56.920 --> 0:05:59.560 a recent data breach that raised questions about how the 0:05:59.600 --> 0:06:03.080 company handle sensitive information and led one of its clients, 0:06:03.240 --> 0:06:07.360 Meta to indefinitely pause its work with the startup. If 0:06:07.440 --> 0:06:10.320 Mercore and its backers are right, the company is uniquely 0:06:10.360 --> 0:06:13.960 positioned to help AI become the economic force. Silicon Valley 0:06:13.960 --> 0:06:16.920 has been promising. Most of what's been seen from AI 0:06:17.080 --> 0:06:19.960 so far, they argue, has been something of a dazzling 0:06:20.040 --> 0:06:25.080 consumer demo. Going beyond that, making the technology perform reliably 0:06:25.160 --> 0:06:29.800 in professional fields where mistakes carry real consequences is fundamentally 0:06:29.800 --> 0:06:33.599 about feeding the models better training data, says Sundeep Peachew, 0:06:33.960 --> 0:06:37.560 managing partner at the VC firm Felicis Ventures, which led 0:06:37.600 --> 0:06:41.440 Mercor's two most recent funding rounds. The first generation of 0:06:41.520 --> 0:06:44.360 data was from the Internet, he says, and that allowed 0:06:44.400 --> 0:06:48.560 companies to build very general purpose models. But for AI 0:06:48.680 --> 0:06:52.280 to become truly economically useful and not just a toy thing, 0:06:52.960 --> 0:06:55.599 someone has to get humans to tell the model, step 0:06:55.640 --> 0:06:59.640 by step how they actually do their work. At this moment, 0:06:59.720 --> 0:07:02.640 it's hard to understand why people are so willing to 0:07:02.680 --> 0:07:05.800 feed the machines that might one day render them obsolete. 0:07:06.560 --> 0:07:10.760 Anxiety about job loss is everywhere. Job openings in professional 0:07:10.800 --> 0:07:13.520 and business services have fallen by more than a million 0:07:13.560 --> 0:07:17.040 from their post pandemic peak in twenty twenty two, according 0:07:17.040 --> 0:07:20.520 to the Bureau of Labor Statistics. About forty two percent 0:07:20.600 --> 0:07:23.880 of recent college grads are under employed, according to a 0:07:23.920 --> 0:07:26.480 twenty twenty five report from the Federal Reserve Bank of 0:07:26.480 --> 0:07:29.320 New York, and a study last year by the American 0:07:29.360 --> 0:07:33.440 Psychological Association found that fifty four percent of US workers 0:07:33.480 --> 0:07:38.960 are experiencing significant stress about job in security. Mercore's listings 0:07:39.000 --> 0:07:42.320 sometimes ads from the company itself and sometimes from third 0:07:42.360 --> 0:07:45.920 parties using Mercore links that allow them to collect referral fees, 0:07:46.440 --> 0:07:50.440 Saturate LinkedIn, and other job boards, offering a respite from 0:07:50.440 --> 0:07:54.000 all that. Mercore says its average hourly pay is about 0:07:54.080 --> 0:07:57.960 ninety dollars, though the range is wide, from generalists barely 0:07:58.000 --> 0:08:02.160 scratching minimum wage to elite coders or PhDs making two 0:08:02.240 --> 0:08:05.520 hundred and fifty or three hundred dollars per hour. Someone 0:08:05.560 --> 0:08:08.600 working on Mercore Project's full time at ninety dollars per 0:08:08.600 --> 0:08:11.520 hour would earn almost one hundred and ninety thousand dollars 0:08:11.520 --> 0:08:14.880 a year. There are plenty of workers who view Mercore 0:08:15.000 --> 0:08:19.040 as a grim final stop to monetize their expertise before 0:08:19.080 --> 0:08:23.360 professional extinction. Kozak doesn't see it that way, saying her 0:08:23.400 --> 0:08:26.120 reason for doing the work is far more practical. She 0:08:26.320 --> 0:08:31.120 envisions herself off loading social work's tedious thirty percent, the paperwork, 0:08:31.160 --> 0:08:34.000 the reports, so she can spend more time doing the 0:08:34.040 --> 0:08:38.320 parts that can't be automated, coaching someone through a bureaucratic maze, 0:08:38.600 --> 0:08:42.240 gaining a grieving mother's trust. She worries about the potential 0:08:42.280 --> 0:08:47.359 for AI to develop systematic biases, say overlooking cultural differences, 0:08:47.800 --> 0:08:50.200 but sees her training as all the more valuable for 0:08:50.240 --> 0:08:54.520 that reason. Doctor Punache is more explicit about the stakes 0:08:54.520 --> 0:08:57.680 for her own career. I am doing this just because 0:08:57.720 --> 0:09:01.440 I don't want to become obsolete, she says. Medicine is 0:09:01.480 --> 0:09:04.760 evolving rapidly, and AI is going to be part of 0:09:04.760 --> 0:09:07.679 that evolution, whether we want to participate in it or not. 0:09:09.280 --> 0:09:13.000 If history has taught us anything about revolutions in productivity, 0:09:13.320 --> 0:09:16.360 it's that productivity is the tide that lifts all boats. 0:09:16.640 --> 0:09:20.720 Brentan foodie, declares one morning in February. The sandy haired, 0:09:20.760 --> 0:09:23.680 twenty three year old co chief executive officer and co 0:09:23.760 --> 0:09:28.160 founder of Mercor has the standard Techno Optimists conversational habit 0:09:28.240 --> 0:09:31.280 of jumping from earthly matters. In this case the field 0:09:31.280 --> 0:09:36.440 of management consulting to historical analogies and abstract principles. Asked 0:09:36.440 --> 0:09:40.320 how a future Mackenzie consultant replaced by AI would gain 0:09:40.400 --> 0:09:44.920 expertise without the traditional entry level drudgery. Foody zooms out 0:09:44.960 --> 0:09:48.720 to a mainstay Silicon Valley position two hundred years ago, 0:09:48.840 --> 0:09:52.480 when most Americans were farmers, the tractor didn't destroy work. 0:09:52.760 --> 0:09:56.320 It pushed people into different kinds of employment. AI will 0:09:56.320 --> 0:09:59.880 do the same, he says, eliminating some tasks, yes, but 0:10:00.200 --> 0:10:05.840 ultimately creating more value, more opportunity, more progress. History offers 0:10:05.840 --> 0:10:08.480 some support for that view, but the path from one 0:10:08.559 --> 0:10:11.480 kind of work to another has rarely been painless or quick. 0:10:12.120 --> 0:10:14.599 We need to cure cancer and solve climate change and 0:10:14.640 --> 0:10:17.719 go to Mars, Foody says, and I think humans will 0:10:17.720 --> 0:10:19.400 work on a lot of those things once we have 0:10:19.480 --> 0:10:22.840 more productivity in accounting or whatever. These more back office 0:10:22.840 --> 0:10:26.960 functions are from where Foodie sits in a glass walled 0:10:27.000 --> 0:10:29.040 conference room on the thirty third floor of a San 0:10:29.080 --> 0:10:33.240 Francisco skyscraper. The bay extends beneath him in miniature, with 0:10:33.360 --> 0:10:37.600 container ships sliding past and ferries carrying commuters to innumerable 0:10:37.600 --> 0:10:41.560 office buildings. From this height, the physical economy looks less 0:10:41.640 --> 0:10:44.559 like the labor of millions of individuals living full lives 0:10:44.920 --> 0:10:47.719 and more like a table model whose tiny pieces could 0:10:47.760 --> 0:10:51.520 be shuffled with mere keystrokes. Instead of being luddites and 0:10:51.600 --> 0:10:54.840 leaning against the technology, we should instead focus on what 0:10:54.920 --> 0:10:56.719 are the jobs of the future that we need to 0:10:56.800 --> 0:11:01.480 lean into, he says. Footy acquires his rhetorical instincts as 0:11:01.520 --> 0:11:04.680 a competitive debater and perpetual hustler in high school in 0:11:04.720 --> 0:11:08.560 San Jose, where he met Hiramat and Meta. All three 0:11:08.640 --> 0:11:12.120 were raised in tech industry households. Hiramot and Meta are 0:11:12.160 --> 0:11:16.160 the sons of Silicon Valley engineers. Foudie's father founded an 0:11:16.200 --> 0:11:19.400 interactive graphics company, and his mother worked in Meta's real 0:11:19.520 --> 0:11:23.480 estate division. As high schoolers, Foodie earned hundreds of thousands 0:11:23.520 --> 0:11:27.640 of dollars as a consultant for sneaker resellers, while Hiramot 0:11:27.640 --> 0:11:31.800 was geeking out on computer vision research. Meta declined to 0:11:31.840 --> 0:11:35.920 speak to Bloomberg BusinessWeek. By their second year of college, 0:11:36.000 --> 0:11:40.000 Hiramat at Harvard University, FOODI, and Meta at Georgetown. The 0:11:40.040 --> 0:11:43.720 three figured they'd learned enough and applied for a Teal Fellowship, 0:11:43.920 --> 0:11:46.760 which pays kids two hundred thousand dollars to leave college 0:11:47.000 --> 0:11:50.720 and start companies in fields such as cryptocurrency and human longevity. 0:11:51.520 --> 0:11:54.000 I think people overestimate the value you get out of 0:11:54.040 --> 0:11:57.040 a four year degree, but underestimate the value you get 0:11:57.080 --> 0:12:00.280 in one or two years. Hiromot says. The bulk of 0:12:00.320 --> 0:12:02.960 the personal development just comes from living on your own 0:12:03.000 --> 0:12:06.720 for the first time. Knowledge acquisition is even less relevant 0:12:06.720 --> 0:12:09.760 today than it was three years ago, he says, because 0:12:09.800 --> 0:12:13.040 knowledge is just kind of free in chat GPT. The 0:12:13.080 --> 0:12:15.120 knowledge that I want I can just obtain with a 0:12:15.120 --> 0:12:18.520 couple of prompts. The three had just begun to zero 0:12:18.600 --> 0:12:21.280 in on AI training when they started the fellowship in 0:12:21.320 --> 0:12:25.040 early twenty twenty four. The company's original idea was a 0:12:25.080 --> 0:12:28.760 far more sweeping version of LinkedIn. The labor market might 0:12:28.800 --> 0:12:32.000 be the most inefficient marketplace in the world. Foody figured, 0:12:32.320 --> 0:12:35.480 with billions of people looking for work, millions of companies 0:12:35.520 --> 0:12:38.920 looking for talent, and no central clearinghouse connecting the two 0:12:39.720 --> 0:12:43.199 mercor would build a kind of global labor aggregator, where 0:12:43.240 --> 0:12:46.679 everyone everywhere could apply, an interview and be vetted for 0:12:46.800 --> 0:12:50.840 every job, and algorithms would match workers to opportunities with 0:12:50.920 --> 0:12:55.080 near perfect precision. In Foody's telling, even the concept of 0:12:55.080 --> 0:12:57.720 a full time job with a single employer was something 0:12:57.720 --> 0:13:01.080 of a relic. What companies actually need needed were discrete 0:13:01.200 --> 0:13:05.280 units of expertise, tasks that could be broken apart, distributed, 0:13:05.320 --> 0:13:07.800 and completed by whoever in the world happened to be 0:13:07.800 --> 0:13:10.840 best suited to them. To make any of that possible 0:13:10.880 --> 0:13:14.520 would require assembling an extraordinary amount of data about people. 0:13:15.600 --> 0:13:18.600 They began with an AI powered tool for screening software 0:13:18.600 --> 0:13:21.520 engineers that served as a prototype for the far bigger 0:13:21.559 --> 0:13:25.640 system they imagined. They changed focus, though, when AI companies 0:13:25.679 --> 0:13:29.160 began to need not only engineers but also other domain 0:13:29.200 --> 0:13:33.040 experts who could help train and test the models. Mercore 0:13:33.120 --> 0:13:36.320 built a tech platform to host the training itself and 0:13:36.440 --> 0:13:41.120 demand exploded. Venture capitalists eager to fund anything tied to 0:13:41.160 --> 0:13:45.720 the AI boom, started calling Mercoor raised three million dollars 0:13:45.760 --> 0:13:49.480 in early twenty twenty four, led by General Catalyst. It 0:13:49.600 --> 0:13:52.000 raised another round every six or eight months after that. 0:13:52.400 --> 0:13:56.040 Thirty million dollars, one hundred million dollars, then three hundred 0:13:56.080 --> 0:14:00.760 and fifty million dollars last October. Footy Here and Meta 0:14:00.880 --> 0:14:04.640 were also working around the clock, encouraging their growing team 0:14:04.760 --> 0:14:07.040 to adhere to a schedule known as nine to nine 0:14:07.160 --> 0:14:10.320 six in the office from nine am to nine pm 0:14:10.440 --> 0:14:14.240 Monday through Saturday. The regiment first became popular in China 0:14:14.280 --> 0:14:16.520 in the twenty tens as the country built out its 0:14:16.559 --> 0:14:22.000 tech industry. It's now illegal there. Felicia's investor Peachu says 0:14:22.040 --> 0:14:24.400 Foodi and his partners were so hard to pin down 0:14:24.800 --> 0:14:27.040 that he eventually flew with them to Las Vegas on 0:14:27.080 --> 0:14:30.080 a Sunday, the one day they weren't working to drive 0:14:30.120 --> 0:14:32.720 ferraris on a racetrack, just so he could get a 0:14:32.720 --> 0:14:36.440 little uninterrupted time with them on the plane. The founders 0:14:36.440 --> 0:14:38.760 weren't old enough to drink at that time, and none 0:14:38.760 --> 0:14:41.200 of them had ever held a professional job, but the 0:14:41.280 --> 0:14:44.120 hundreds of millions of dollars being plowed into their coffers 0:14:44.200 --> 0:14:47.080 came with a mandate to remake the very nature of 0:14:47.120 --> 0:14:50.880 white collar work in Silicon Valley. That didn't raise any 0:14:50.920 --> 0:14:56.160 alarms about naivete or blind spots. Quite the contrary, Adam DeAngelo, 0:14:56.480 --> 0:14:59.800 a Mercor investor who became Facebook's chief technology officer at 0:14:59.880 --> 0:15:03.760 Day twenty two before co founding Qua, says he asked 0:15:03.760 --> 0:15:07.160 Foody at one point about his work experience. FOODI said 0:15:07.200 --> 0:15:11.520 he'd had an internship one summer perfect. Dangelo said, your 0:15:11.560 --> 0:15:14.720 mind isn't corrupted by the conventional way of doing things. 0:15:16.360 --> 0:15:19.120 We'll be right back with the ten billion dollars start 0:15:19.200 --> 0:15:25.480 up training AI to replace the white collar workforce. Welcome 0:15:25.520 --> 0:15:28.480 back to the ten billion dollars start up training AI 0:15:28.640 --> 0:15:32.800 to replace the white collar workforce. Over the past year, 0:15:32.960 --> 0:15:36.880 criticism of Mercor's platform has become a refrain in online forums, 0:15:37.200 --> 0:15:41.120 news reports, and court filings, Beginning as these things often 0:15:41.160 --> 0:15:46.040 do with anonymous online posts. Reddit in particular contains no 0:15:46.160 --> 0:15:49.880 shortage of grievances from contractors who've cycled through the platform. 0:15:50.440 --> 0:15:54.920 Projects are chaotic, disorganized, and unpredictable. Workers are treated like 0:15:55.040 --> 0:15:59.000 human cattle. The company is building the plane while flying. 0:16:00.080 --> 0:16:04.560 Contractors signed strict confidentiality agreements before beginning work, and many 0:16:04.640 --> 0:16:07.640 fear being removed from its projects if they speak publicly. 0:16:08.240 --> 0:16:11.400 They gripe about slack channels full of motivational chatter and 0:16:11.480 --> 0:16:15.360 rocket ship emoji from project managers, but few concrete answers 0:16:15.360 --> 0:16:19.400 about when work will arrive or advance notice when it stops. 0:16:19.440 --> 0:16:22.760 Like many forms of digital piece work, Mercor's projects are 0:16:22.760 --> 0:16:26.600 closely measured and monitored while they're underway, with software tracking 0:16:26.640 --> 0:16:30.280 productivity and time spent on each assignment. It turns out 0:16:30.320 --> 0:16:33.400 some jobs pay not by the hour, but by the task. 0:16:33.920 --> 0:16:37.720 A frustrated contractor BusinessWeek spoke with, who asked for anonymity 0:16:37.760 --> 0:16:41.080 because he'd signed what a mercoor's non disclosure agreements, has 0:16:41.120 --> 0:16:43.240 been job hunting for almost a year for a full 0:16:43.280 --> 0:16:46.400 time position to make use of his master's degree in physics. 0:16:46.960 --> 0:16:50.400 After he joined Mercoor. His first contract paid thirty dollars 0:16:50.440 --> 0:16:53.200 per task, but he soon learned that the time it 0:16:53.200 --> 0:16:56.040 took for him to complete tasks anywhere from an hour 0:16:56.160 --> 0:16:58.840 to a full day, was often far greater than the 0:16:58.840 --> 0:17:02.640 company's estimates, making the pay too low to justify the work. 0:17:03.240 --> 0:17:06.159 It's a variation of a complaint that services online about 0:17:06.160 --> 0:17:10.359 hourly projects too. Contractors sometimes quietly do part of their 0:17:10.400 --> 0:17:13.159 work off the clock to keep their productivity numbers in 0:17:13.200 --> 0:17:16.679 line with company benchmarks, out of fear of being off boarded. 0:17:17.119 --> 0:17:21.200 Mercy speak for getting booted. A mercoor spokesperson says that 0:17:21.400 --> 0:17:26.200 many experienced contractors prefer task based work because increased efficiency 0:17:26.200 --> 0:17:30.000 can lead to higher effective hourly earnings. Last fall, one 0:17:30.080 --> 0:17:33.840 Mercor contractor dispute spilled into the press when Forbes reported 0:17:33.880 --> 0:17:36.560 that thousands of people working on a large project were 0:17:36.560 --> 0:17:40.840 abruptly locked out without warning. Several hours later, some said 0:17:41.080 --> 0:17:43.920 they were invited back to continue the work, but at 0:17:43.920 --> 0:17:48.040 pay rates roughly a quarter lower than before. Mercore disputed 0:17:48.040 --> 0:17:50.520 the accuracy of the claims and said it was working 0:17:50.520 --> 0:17:53.959 to offer more predictability. The company also alleges that some 0:17:54.000 --> 0:17:56.320 of the people who complain about being let go are 0:17:56.320 --> 0:17:59.800 simply liars. We found cases where people have been doing 0:17:59.840 --> 0:18:03.159 no no work or doing time fraud. Hiramat says, we 0:18:03.240 --> 0:18:06.240 have demonstrable evidence of them committing fraud, but they still 0:18:06.280 --> 0:18:09.040 post about it on Reddit when their contract has been eliminated. 0:18:09.960 --> 0:18:13.960 As criticism of Mercore has grown, lawsuits have followed. One 0:18:13.960 --> 0:18:16.480 of the complaints filed late last year by a finance 0:18:16.520 --> 0:18:20.679 professional named Michael Cox, who worked on Mercore projects, lists 0:18:20.720 --> 0:18:23.920 open Ai as a co defendant and accuses the companies 0:18:23.960 --> 0:18:27.040 of running what it calls a scheme to misclassify workers 0:18:27.440 --> 0:18:31.360 while exercising the kind of control normally associated with an employer. 0:18:31.960 --> 0:18:35.240 According to the filing, Mercoy required Cox to install a 0:18:35.240 --> 0:18:38.919 productivity monitoring software on his computer, creating a level of 0:18:38.960 --> 0:18:42.520 surveillance so intrusive as to trivialize the notion that this 0:18:42.640 --> 0:18:47.119 was a legitimate independent contractor relationship. In another passage, the 0:18:47.160 --> 0:18:50.480 complaint says the company's alleged neglect of employment norms was 0:18:50.560 --> 0:18:55.160 so brazen as to almost by definition, constitute wilful misclassification. 0:18:55.920 --> 0:18:59.399 Mercore and OpenAI haven't publicly responded to the suit, and 0:18:59.440 --> 0:19:02.119 a Mercoors spokesperson says that they don't plan to at 0:19:02.160 --> 0:19:08.800 this time. Two other suits against mercor make similar misclassification claims. Meanwhile, 0:19:08.960 --> 0:19:11.320 if you were a job seeker on LinkedIn last fall, 0:19:11.640 --> 0:19:15.639 Mercy suddenly seemed to be everywhere. Listings offering hundreds of 0:19:15.640 --> 0:19:19.560 dollars an hour to lawyers, doctors, and programmers flooded the platform, 0:19:19.720 --> 0:19:23.680 prompting another wave of social media speculation, this one less 0:19:23.680 --> 0:19:26.920 about working conditions than about whether the whole thing was real. 0:19:27.600 --> 0:19:30.960 Some users began posting on various sites speculating that the 0:19:31.000 --> 0:19:34.640 listings were an elaborate data harvesting scheme in which fake 0:19:34.680 --> 0:19:39.480 applications and AI interviews were designed to collect valuable personal information. 0:19:40.200 --> 0:19:42.679 Mercoor says that this isn't the case, and that it 0:19:42.800 --> 0:19:46.119 uses interviews only to evaluate candidate's skills for jobs on 0:19:46.160 --> 0:19:50.679 the platform, and again the company suggests its critics are 0:19:50.720 --> 0:19:54.760 the scammers. Foodie says the LinkedIn deluge wasn't about Mercoor 0:19:54.880 --> 0:19:58.600 harvesting data, but about a few high volume fraudsters harvesting 0:19:58.640 --> 0:20:02.320 referral fees because it pays users to bring new contractors 0:20:02.320 --> 0:20:06.119 into the fold. There were probably ten people specifically that 0:20:06.160 --> 0:20:10.920 were causing problems, he says. In October, Mercoor banned refers 0:20:10.960 --> 0:20:13.480 from using the name mercor when they post job ads 0:20:13.480 --> 0:20:17.280 for the roles on LinkedIn. Still, postings from third party 0:20:17.320 --> 0:20:21.399 referral outfits have continued. Mercore listings from one such organization, 0:20:21.760 --> 0:20:24.800 a recruiter called Crossing Hurdles, based on the outskirts of 0:20:24.840 --> 0:20:28.760 New Delhi, are a constant presence on LinkedIn. A mercor 0:20:28.880 --> 0:20:32.879 spokesperson tells BusinessWeek that Crossing Hurdles has no official affiliation 0:20:33.080 --> 0:20:36.720 or partnership with Mercy. They post roles on various job 0:20:36.760 --> 0:20:41.000 sites using their Mercore referral link crossing hurdles didn't respond 0:20:41.000 --> 0:20:45.560 to multiple requests for comment. Mercor made another change last fall. 0:20:45.880 --> 0:20:48.920 It moved its third co founder, Meta into a new 0:20:48.960 --> 0:20:52.800 position Chairman of the board, handing his previous role running 0:20:52.840 --> 0:20:57.040 operations to a more seasoned executive, Sundeep Jain, a former 0:20:57.119 --> 0:21:01.480 chief product officer at Uber Technologies. In evoking his previous employer, 0:21:01.840 --> 0:21:05.359 Jane says that the complaints circulating online aren't surprising and 0:21:05.400 --> 0:21:09.520 that early stage marketplaces rarely distribute work evenly. There will 0:21:09.560 --> 0:21:11.959 be some drivers that will be busy all day and 0:21:12.080 --> 0:21:14.560 others that'll be a little bit less busy, he says. 0:21:15.080 --> 0:21:20.760 Communication hiccups, payment disputes, mismatched expectations, those are classic problems 0:21:20.760 --> 0:21:24.640 of a marketplace. But on March thirty, first, the young company, 0:21:25.000 --> 0:21:27.560 growing at breakneck speed and reliant on a handful of 0:21:27.640 --> 0:21:31.640 high profile clients experienced something far worse than a hiccup. 0:21:32.119 --> 0:21:35.440 Mercor disclosed that attackers had infiltrated its systems in a 0:21:35.480 --> 0:21:39.320 sprawling supply chain hack. The breach coming through a corrupted 0:21:39.359 --> 0:21:43.159 open source developer tool called light LLM that thousands of 0:21:43.160 --> 0:21:46.919 companies use. Struck at the core of Mercor's business. As 0:21:47.000 --> 0:21:50.600 much as four terabytes of data, including possibly training data 0:21:50.640 --> 0:21:55.040 and user's personal information, were exposed, according to online posts 0:21:55.040 --> 0:21:58.119 from the hackers. The company announced it was conducting a 0:21:58.200 --> 0:22:02.639 thorough investigation, supported by leading third party forensics experts, but 0:22:02.720 --> 0:22:06.360 the fallout was swift. Along with Meta pausing its projects 0:22:06.359 --> 0:22:10.200 with Mercor, contractors assigned to those projects suddenly found themselves 0:22:10.200 --> 0:22:13.240 without work. In the first week after announcing the breach, 0:22:13.560 --> 0:22:16.520 Mercor was hit with five lawsuits accusing it of failing 0:22:16.560 --> 0:22:21.000 to protect contractor's data. Mercore declined to respond to BusinessWeek's 0:22:21.080 --> 0:22:23.679 questions about the origin and extent of the hack and 0:22:23.760 --> 0:22:27.600 its impact on clients and users. Open Ai Andthropic and 0:22:27.640 --> 0:22:30.960 Google didn't respond to requests for comment on whether they're 0:22:30.960 --> 0:22:35.000 now re evaluating their relationship with Mercor. Beyond the messy 0:22:35.119 --> 0:22:38.480 day to day mechanics of atomizing jobs into training tasks 0:22:38.880 --> 0:22:42.240 is a bigger question. Are the machines actually getting good 0:22:42.320 --> 0:22:45.400 enough to do the work for themselves. In the past year, 0:22:45.480 --> 0:22:48.639 the AI industry has tried to answer that question, mostly 0:22:48.680 --> 0:22:53.800 with a growing ecosystem of evals or evaluation frameworks like 0:22:53.840 --> 0:22:57.440 a standardized test. An eval measures a large language model's 0:22:57.480 --> 0:23:01.879 math ability, reasoning, or factual accuisi, but not typically the 0:23:01.960 --> 0:23:05.760 kinds of specific, judgment heavy tasks that white collar workers 0:23:05.800 --> 0:23:09.399 do all day. Mrcore has developed its own test, a 0:23:09.440 --> 0:23:13.959 project called apex AI Productivity Index that attempts to do 0:23:14.200 --> 0:23:17.639 just that. It measures professional performance and then posts the 0:23:17.680 --> 0:23:20.000 results on the web for anyone to follow as a 0:23:20.080 --> 0:23:23.280 kind of industry leader board. As Foodie put it when 0:23:23.280 --> 0:23:27.400 announcing the initiative last October, AI can pass the bar exam, 0:23:27.840 --> 0:23:32.040 but can it redline a contract? To build apex murcoor 0:23:32.119 --> 0:23:35.040 has worked with some of its most accomplished contractors to 0:23:35.119 --> 0:23:43.480 design short professional scenarios and problems in five fields so far, law, medicine, management, consulting, investment, banking, 0:23:43.520 --> 0:23:48.200 and software engineering. Advisors overseeing the effort include Harvard Law 0:23:48.200 --> 0:23:53.119 professor Cass Sunstein, cardiologist Eric Topel, and former McKenzie Global 0:23:53.160 --> 0:23:57.840 managing partner Dominic Barton. Former Treasury secretary and Mercore investor 0:23:58.000 --> 0:24:02.080 Larry Summers was until recently also affiliated with the project. 0:24:02.680 --> 0:24:05.359 Beneath them is a larger bench of experts, more than 0:24:05.400 --> 0:24:09.159 one hundred lawyers, bankers, consultants and clinicians, who build and 0:24:09.240 --> 0:24:12.680 review the actual test cases, the models do the work, 0:24:12.920 --> 0:24:16.280 and their responses are graded against expert benchmarks to see 0:24:16.280 --> 0:24:20.320 how close they come to professional quality. One case, for instance, 0:24:20.520 --> 0:24:24.760 centers on a fictional wellness company expanding overseas. A Google 0:24:24.800 --> 0:24:28.359 workspace contains the kind of unwieldy digital paper trail a 0:24:28.440 --> 0:24:31.720 junior consultant might inherit on the first day of a project, 0:24:32.160 --> 0:24:36.720 sales data customer surveys, cost projections, and strategy memos scattered 0:24:36.760 --> 0:24:41.399 across spreadsheets and PDFs. Mercor's experts prompt AI models to 0:24:41.400 --> 0:24:45.840 answer strategic questions such as calculating how rising ingredient costs 0:24:45.920 --> 0:24:49.840 might affect pricing or recommending an expansion strategy, and define 0:24:49.880 --> 0:24:52.760