Datacurve raises $15 million to take on ScaleAI

As AI companies mature, the combat for high-quality information has go 1 of the astir competitory areas successful the industry, launching companies similar Mercor, Surge, and, astir prominently, Alexandr Wang’s ScaleAI. But present that Wang has moved to tally AI astatine Meta, galore funders spot an opening — and are consenting to money companies with compelling caller strategies for collecting grooming data.

The Y Combinator postgraduate Datacurve is 1 specified company, focusing connected high-quality information for bundle development. On Thursday, the institution announced a $15 cardinal Series A round, led by Mark Goldberg astatine Chemistry with information from employees astatine DeepMind, Vercel, Anthropic, and OpenAI. The Series A comes aft a $2.7 cardinal effect round, which drew concern from erstwhile Coinbase CTO Balaji Srinivasan.

Datacurve uses a “bounty hunter” strategy to pull skilled bundle engineers to implicit the hardest-to-source datasets. The institution pays for those contributions, distributing implicit $1 cardinal successful bounties truthful far.

But co-founder Serena Ge says the biggest information isn’t financial. For high-value services similar bundle development, the wage volition ever beryllium acold little for information enactment than accepted employment — truthful the company’s astir important borderline is simply a affirmative idiosyncratic experience.

“We dainty this arsenic a user product, not a information labeling operation,” Ge said. “We walk a batch of clip reasoning about: How tin we optimize it truthful that the radical we privation are funny and get onto our platform?”

That’s peculiarly important arsenic the needs of post-training information turn much complex. While earlier models were trained connected elemental datasets, today’s AI products trust connected complex RL environments, which request to beryllium constructed done circumstantial and strategical information collection. As the environments turn much sophisticated, the information requirements go some much aggravated for some quantity and prime — a origin that could springiness high-quality information postulation companies similar Datacurve an edge.

As an early-stage company, Datacurve is focused connected bundle engineering, but Ge says the exemplary could use conscionable arsenic easy to fields similar finance, marketing, oregon adjacent medicine.

“What we’re doing close present is we’re creating an infrastructure for station grooming information postulation that attracts and retains highly competent radical successful their ain domains,” Ge says.