Despite growing discussions about a future when most human tasks are automated by AI, one irony of this current technological boom is how stubbornly dependent it is on humans, particularly the process of training AI models using reinforcement learning from human feedback (RLHF).
In its simplest form, RLHF is a learning system: even after training an AI on curated data, it still makes mistakes or seems robotic. Human contractors are collectively hired by AI laboratories to rate and rank the output of a new model during training, and the model learns from their ratings, adjusting its behavior to offer higher-rated outputs. This process is even more important as AI expands to produce multimedia outputs such as video, audio, and imagery that can have more nuanced and subjective measures of quality.
Historically, this tuition process has been a huge logistical headache and PR nightmare for AI companies, relying on fragmented networks of overseas contractors and static labeling pools in specific, low-income geographic centres, perceived by the media as low-wage – even exploitative. It’s also inefficient: AI labs have to wait weeks or months for a batch of feedback, delaying model progress.
Now a new startup has emerged to make this process more efficient: Rapidata’s platform effectively "gamifies" RLHF has pushed the said review tasks to approximately 20 million users of popular apps worldwide, including Duolingo or Candy Crush, as small, opt-in review tasks that they can choose to complete in place of viewing a mobile ad, with the data immediately fed back to the commissioning AI lab.
As shared with VentureBeat in a press release, the platform allows AI labs "iterate over models in near real time," Development timelines are significantly reduced compared to traditional methods.
CEO and founder Jason Corkill said in the same release that Rapidata makes "Human decision-making, available on a global scale and in near real-time, is opening up a future where AI teams can drive continuous feedback loops and create systems that evolve every day, rather than every release cycle.""
Rapidata treats RLHF as a high-speed infrastructure rather than a manual labor problem. Today, the company exclusively told us at VentureBeat announcing its emergence with an $8.5 million seed round co-led by Canaan Partners and IA Ventures, with participation from Acequia Capital and Blueyard, to scale its unique approach to on-demand human data.
The pub conversation that created the human cloud
Rapidata was born not in a boardroom, but around a table with a few beers. Corkill was a student at ETH Zurich, working in robotics and computer vision, when he hit the wall that every AI engineer eventually faces: the data annotation barrier.
"In particular, I have been working in robotics, AI and computer vision for the past few years, having studied at ETH here in Zurich, and was always frustrated with data annotation," Corkill recalled in a recent interview. "Always when you need a human or human data annotation, it’s like the time when your project stopped in its tracks, because till then, you could only push it forward by spending long nights. But when you need human annotation on a large scale, you have to go to someone and then wait for a few weeks.".
Frustrated by this delay, Corkill and his co-founders realized that the existing labor model for AI was fundamentally broken for a world moving at the speed of modern computation. While calculating faster scale, the traditional human workforce – tied to manual onboarding, regional hiring and slow payment cycles – does not. RapidData was born from the idea that human judgment could be delivered as a globally distributed, nearly instantaneous service.
Technology: turning digital footprints into training data
Rapidata’s main innovation lies in its delivery method. Instead of hiring full-time annotators in specific areas, the Rapidata mobile app takes advantage of the world’s existing attention economy. By partnering with third-party apps like Candy Crush or Duolingo, Rapidata offers users a choice: watch a traditional ad or spend a few seconds providing feedback to an AI model.
"Users are asked, ‘Hey, would you like to annotate some data, give feedback, instead of watching ads and, you know, companies buying your eyeballs?’" Corkill explained. According to Corkill, 50% to 60% of users choose feedback engagement over traditional video advertising.
it "crowd intelligence" The approach allows AI teams to tap into diverse, global demographics on an unprecedented scale.
- global network: Rapidata currently reaches 15 to 20 million people.
-
Broad similarity: The platform can process 1.5 million human annotations an hour.
-
pace: Feedback cycles that previously took weeks or months are now reduced to hours or minutes.
-
Quality Control: The platform builds trust and expertise profiles for respondents over time, ensuring that complex questions are matched with the most relevant human judges.
-
Anonymity: While users are tracked via anonymized IDs to ensure consistency and reliability, Rapidata does not collect personal identification, maintaining privacy while optimizing data quality.
Online RLHF: Moving to GPU
The most significant technological leap is enabling RapidData which Corkill describes as "online rlhf". Traditionally, AI is trained in disconnected batches: you train the model, stop, send the data to humans, wait weeks for labels, and then start again. it makes a "circle" Information that often lacks fresh human input.
Rapidata is taking this decision directly into the training loop. Because their networks are so fast, they can integrate directly with GPUs running models through APIs.
"We’ve always had this idea of reinforcement learning for human response… Till now, you always had to do it in batches," Corkill said. "Now, if you go all the way down, we have some clients now, where, because we’re so fast, we can be directly, seamlessly into the process, like the GPU right in the processor, and the GPU computes some output, and it can immediately request it to us in a distributed manner. ‘Oh, I want, I want, I need a human being to see it.’ I find the answer and then inflict damage that has not been possible until now".
Currently, the platform supports approximately 5,500 humans per minute providing live feedback to models running on thousands of GPUs. it prevents "reward model hacking," Where two AI models deceive each other in a feedback loop by basing the training on actual human nuances.
Products: solutions for taste and global context
As AI moves beyond simple object recognition to generic media, data labeling requirements have evolved from objective tagging to subjective "taste based" Duration. It’s not just talk anymore "Is it a cat?" Rather "Is this sound synthesis reliable?" Or "Which of these two summaries sounds more professional?".
Lily Clifford, CEO of voice AI startup Rhyme, says RapidData has been transformative for testing models in a real-world context. "Previously, gathering meaningful feedback meant piecing together vendors and surveys, segment by segment, or country by country, with no scale." Clifford said. Using RapidData, Rhyme can reach the right audience – whether it’s in Sweden, Serbia, or the United States – and see how models perform in real customer workflows in days, not months.
"Most models are factually correct, but I’m sure you’ve received emails that, you know, aren’t authentic, right?" Corkill noted. "You can sniff out an AI email, you can sniff out an AI image or video, it’s immediately obvious to you… these models still don’t feel human, and you need human feedback to do that".
Economic and operational changes
From an operational perspective, Rapidata positions itself as an infrastructure layer that eliminates the need for companies to manage their own custom annotation operations. By providing a scalable network, the company is lowering the barrier to entry for AI teams that previously struggled with the cost and complexity of traditional feedback loops.
Jared Newman of Canaan Partners, who led the investment, suggests that this infrastructure is essential for the next generation of AI. "Every serious AI deployment depends on human judgment somewhere in the lifecycle," Newman said. "As models move from expertise-based tasks to taste-based curation, the demand for scalable human feedback will increase dramatically".
future of human use
While the current focus is on the Bay Area’s model labs, Corkill sees a future where AI models themselves will become the primary clients of human judgment. he calls it "human use".
In this view, a car designer AI would not just design a simple vehicle; It can programmatically call Rapidata to ask 25,000 people in the French market what they think about a specific aesthetic, iterate on that feedback, and refine its design within hours.
"Society is constantly changing," Corkill said, addressing the trend of using AI to simulate human behavior. "If they simulate a society now, the simulation will remain stable and maybe mirror ours for a few months, but then it changes completely, because the society has changed and developed in a completely different way".
By creating a distributed, programmatic way to access human brain capacity around the world, RapidData is positioning itself as a critical intersection between silicon and society. With $8.5 million in new funding, the company plans to move aggressively to ensure that as AI scales, the human element is no longer a hindrance, but a real-time facilitator.
<a href