More than ever, we need a comeback episode of @BG2Pod with @bgurley and @altcap sorting out recent events - bringing reason to the current developments around Jensen and geopolitics.
Congrats to the Mercor team for this incredible achievement! Been thinking a lot about this market, and I think that behind this announcement are two major pivots and one big bet about the future of AI.
First, the pivots.
Mercor started as a talent marketplace, finding experts in coding, health, finance, law, etc. to provide raw data for the frontier labs' training pipelines. That drove eye popping growth for Mercor and its cousins Scale, Surge, et al in 2024 and 2025.
Fast forward to today, and it's no longer fair to call these "staffing" or "data labeling" companies. Most have essentially become research shops. They work with the frontier labs to figure out which areas matter most, find experts in them, and build the RL environments and evals that become direct inputs to model training. They sell that "research" to the labs, who train on it and, in turn, sell their models to enterprises and consumers, token by token.
That's pivot #1. They used to sell the steel. Now they sell the car parts, and the labs build the car.
Pivot #2 is serving enterprises. There aren't that many frontier labs, and unless you're a monopoly supplier of a supply-constrained good (eg TSMC to NVIDIA & Apple), you're gonna have a bad time serving an oligopoly. Just ask Scale, which lost Google and OpenAI as customers overnight after the Meta deal. A diverse market means no single buyer has that leverage. But that market only exists if companies actually want to own their models. 🙃
Which leads me to the big bet: companies will "own their own intelligence" instead of renting it from OpenAI or Anthropic, post-training open weight models with Mercor's research expertise (+expert marketplace). Intelligence gets federated outwards instead of owned by a few large firms, and "frontier" moves to the edges: the companies delivering services in each domain.
Honestly, I'm excited by that future. Even more so if it went all the way down to individuals -- each person owning an intelligence tuned to how they think, work & contribute to society. 8B+ separate intelligences instead of one giant global brain doing all the thinking that will ever be done.
But as I watch new model adoption and the seemingly unstoppable benefits of scaling laws, I wonder if that future is just my fantasy. We don't seem anywhere near their bounds, and the curve still greatly rewards horizontal data aggregators: a model well trained in law tends to also outperform in business strategy. That gives labs a capital advantage on compute & data that no single company can match. When does this change? When do the labs see diminishing returns? Will they ever?
If they do, companies like Mercor have an unlimited market: the knowledge needs of the entire world economy.
If they don't, Mercor & co will still have a business providing research services to labs, but will always live in their shadow -- and at their mercy.
It could be vastly underestimated how much human expertise is needed to diffuse the benefits of AI to the long-tail of knowledge work. I think more data companies should figure out this branding, where experts help AI companies understand how to improve the tools that make them more effective, rather than replaced.
Some jobs will be replaced for sure, but I think a lot of knowledge work is deeply interpersonal, messy, and context driven in a way where AI adoption will be a productivity boon to existing employees. We are seeing this with the demand for SWEs today.
This rebrand will serve the AI industry writ large very well, but goes against the RSI narratives of the labs as "ASI will replace all knowledge work." In some ways, the difference is more philosophical about "what is the nature of work." I believe knowledge work is much more human than the evals give credit for, and this will make diffusion take a long time. Seems like Mercor can be at the center of this diffusion puzzle.
Disclosure: Yes, I advise Mercor on trying to realize this puzzle, as I think it's super interesting.
Thx for the post @hannah . We at @redpineai get paid for quality and professionalism - in data, process, and speed. Quality data translates into lower pass rates on tasks in environments. Bar increases quickly but the ceiling is no where near.
The quick growth of the ’data for ai’ space has led to lots of one person shops without quality and professionalism. Labs require and deserve better.
Andreas Almgren is a role model for athletes of all ages. At tonight’s inspiration talk at Verdane in Stockholm, he shared the recipe for success. A week ago, he won the NYC marathon.
Andreas is working with data on everything from his runs, his races, competitors, tracks, nutrition is at the heart of his method.
After his injury years back, he switched the strategy towards higher training dosage but not pushing himself over the limit every training.
By having experienced injury and failure, he puts less stress on himself before races, and tries to enjoy it instead.
Having the mental strength to push through and welcome pain and negative thoughts, and turn them around to positive fuel.
Many of these lessons applies equally well towards entrepreneurship. Building is a marathon - but in high pace. Using data, accepting risk to win, and to apply enormous grit are key principles.
Thanks team Verdane (Caroline, Tim and team) for this insightful event!
A few years ago, I was at a lab training models and frustrated with data quality. Now I run a data research lab where former colleagues are buyers. It has been a humbling, fascinating and incredibly rewarding (no pun intended) adventure.
Some of what I’ve learned
@termsheetinator Thanks, good nuggets and tips. Building trust in meeting one is the key factor, including how to ensure data is safe during transfer, answer factually about ensuring no reproduction, PII removals and making sure competitors never get access.
Special evening tonight. Had a brief chat with Jensen Huang in Stockholm about the importance of building the knowledge layer of AI. He also added that knowledge graphs were important for structure and retrieval.
As a small token of appreciation, I gave him a Redpine workout t-shirt, for both pre-training and post-training use.
Jensen was in town to open a new AI Center, and the Sferical Labs super computer data center, in collaboration with the Wallenberg Family. His presence and energy was simply unique.
He delivered a brilliant address, recognising many of the companies in the ecosystem, including Ericsson, Spotify and more recent firms like Lovable and Polar.
Finally, to a room of 200+ entrepreneurs and investors, he encouraged Europe to keep building all layers of the AI stack.