Rapidata Aims to Turn Mobile Games Into RLHF Engine

AI systems may be getting more capable, but they are still heavily dependent on human help to learn how to behave. A new startup called Rapidata is stepping into that gap with a platform that aims to shorten AI model development cycles from months to days by rethinking how reinforcement learning from human feedback (RLHF) is done.

RLHF is a core technique behind today’s leading AI models. After an AI system is trained on large, curated datasets, it often still produces awkward, low‑quality or unreliable outputs. To refine it, AI labs hire human workers to rate and rank the model’s responses. The model then adjusts its behaviour to maximize the kinds of outputs humans prefer.

This “tutoring” step has grown even more important as AI expands beyond text into video, audio and images, where quality can be more subjective and context‑dependent. Subtle judgments about tone, relevance or visual appeal are difficult to capture with automated metrics alone, keeping humans firmly in the loop.

Traditionally, RLHF has been a slow, fragmented and often controversial process. AI companies have relied on networks of contractors in specific, lower‑income regions to label and review model outputs. Media reports have highlighted low pay and difficult working conditions in some of these hubs, adding reputational pressure on top of logistical complexity.

The workflow itself is also inefficient. AI labs may have to wait weeks or even months for a batch of feedback from static labelling pools before they can meaningfully update a model. That delay can hold back progress in a field that is otherwise moving quickly.

Rapidata’s platform proposes a different path. Instead of concentrating RLHF work in dedicated labelling centres, it “gamifies” the process and pushes it out to a much broader audience via popular consumer apps. According to details shared with VentureBeat in a press release, Rapidata can route review tasks to nearly 20 million users of apps such as Duolingo or Candy Crush.

These tasks appear as short, opt‑in micro‑activities that users can choose to complete instead of watching mobile ads. In effect, the time that might have been spent passively consuming an advertisement is redirected toward ranking or reviewing AI outputs.

Once those micro‑tasks are completed, the feedback data is sent back instantly to the AI lab that commissioned it. Rapidata says this near real‑time loop lets labs iterate on models far more quickly than with traditional RLHF setups, potentially compressing development timelines substantially.

Gamified RLHF and the road ahead

By embedding RLHF work into familiar apps and offering it as an alternative to ads, Rapidata is betting that a distributed, opt‑in crowd can provide higher‑volume and faster human feedback than conventional contractor pools. The company positions this as a way for AI builders to refine models more continuously, particularly as they push into richer media formats that demand nuanced human judgment.

CEO and founder Jason Cor is leading the effort to commercialize this approach, with the company presenting its platform as infrastructure for AI labs that want to move from slow, batch‑based feedback cycles to something closer to live tuning.

Rapidata’s emergence underscores an ongoing paradox in AI: even as the industry talks about automating more forms of work, progress at the cutting edge still depends on large numbers of people quietly teaching systems what “good” outputs look like. Whether distributing that work across mobile app users will change the economics, speed or public perception of RLHF remains to be seen, but it points to a future where everyday digital interactions are increasingly entwined with how AI systems learn.