By admin 
Despite the growing discourse surrounding the automation of human labor by artificial intelligence, the current technological boom paradoxically hinges on human involvement, particularly in the crucial process of training AI models through reinforcement learning from human feedback (RLHF). RLHF, at its core, functions as a sophisticated tutoring system. After an AI has been initially trained on curated datasets, it often exhibits flaws or produces outputs that sound unnatural or robotic. To address this, AI laboratories enlist human contractors on a massive scale to evaluate and rank the model’s outputs during its training phases. The AI then learns from these rankings, adjusting its behavior to generate more highly-rated responses. This feedback loop is becoming increasingly vital as AI capabilities expand into generating multimedia content such as video, audio, and imagery, where the assessment of quality is inherently more nuanced and subjective.
Historically, this essential human tutoring process has presented significant logistical challenges and has been a source of public relations difficulties for AI companies. These operations often relied on fragmented networks of contractors located in specific, often low-income, geographic hubs. Media reports have frequently characterized these workers as being subjected to low wages, bordering on exploitative, raising ethical concerns. Furthermore, the traditional model is inefficient; AI labs frequently experience delays of weeks or even months waiting for a single batch of human feedback, significantly hindering the pace of model development and iteration.
Enter Rapidata, a new startup poised to revolutionize this landscape by dramatically enhancing the efficiency of the RLHF process. The company’s innovative platform effectively "gamifies" RLHF by distributing review tasks globally to an expansive network of nearly 20 million users of popular mobile applications, including widely used platforms like Duolingo and Candy Crush. These tasks are presented as short, optional review prompts that users can choose to complete as an alternative to watching mobile advertisements. The crucial advantage is the near-instantaneous transmission of this feedback data back to the commissioning AI lab.
In a press release shared with VentureBeat, Rapidata highlighted how its platform empowers AI laboratories to "iterate on models in near-real-time," a significant departure from traditional methods that resulted in considerably longer development timelines. Jason Corkill, CEO and founder of Rapidata, emphasized this transformative capability, stating in the same release that Rapidata makes "human judgment available at a global scale and near real time, unlocking a future where AI teams can run constant feedback loops and build systems that evolve every day instead of every release cycle."
The genesis of Rapidata, as revealed by Corkill in a recent interview, was not conceived in a corporate boardroom but rather emerged from a more informal setting – a pub conversation among friends. While a student at ETH Zurich, deeply involved in robotics and computer vision, Corkill repeatedly encountered the formidable data annotation bottleneck that is a common hurdle for AI engineers. "Specifically, I’ve been working in robotics, AI and computer vision for quite a few years now, studied at ETH here in Zurich, and just always was frustrated with data annotation," Corkill recalled. "Always when you needed humans or human data annotation, that’s kind of when your project was stopped in its tracks, because up until then, you could move it forward by just pushing longer nights. But when you needed the large scale human annotation, you had to go to someone and then wait for a few weeks." This pervasive frustration with the delays inherent in traditional annotation processes fueled the co-founders’ realization that the existing labor model for AI was fundamentally misaligned with the rapid pace of modern technological advancement. While computational power scales exponentially, the traditional human workforce, constrained by manual onboarding processes, geographically concentrated hiring, and slow payment cycles, simply could not keep pace. Rapidata was thus born from a core idea: that human judgment could be delivered as a globally distributed, near-instantaneous service.
The company’s core technological innovation lies in its novel distribution method for collecting feedback. Instead of the conventional approach of hiring dedicated annotators in specific geographical regions, Rapidata ingeniously harnesses the existing attention economy prevalent in the mobile app ecosystem. By forging partnerships with third-party applications such as Candy Crush and Duolingo, Rapidata offers users a compelling choice: they can opt to watch a traditional video advertisement, or they can dedicate a few seconds to providing valuable feedback for an AI model. "The users are asked, ‘Hey, would you rather instead of watching ads and having, you know, companies buy your eyeballs like that, would you rather like annotate some data, give feedback?’" Corkill explained. The results have been striking, with Corkill reporting that a significant majority of users, between 50% and 60%, choose the feedback task over a conventional video advertisement. This "crowd intelligence" approach provides AI development teams with unprecedented access to a diverse, global demographic for data annotation.
Rapidata is enabling a significant technological leap, which Corkill describes as "online RLHF." Traditionally, AI training occurs in discrete batches: the model is trained, the process halts, data is sent to human evaluators, labels are awaited for weeks, and only then does training resume. This creates an information "circle" that often lacks the immediacy of fresh human input. Rapidata, however, is integrating human judgment directly into the training loop. Due to the exceptional speed of their feedback network, Rapidata can integrate via API directly with the graphics processing units (GPUs) that are running the AI models. "We’ve always had this idea of reinforcement learning for human feedback… so far, you always had to do it like in batches," Corkill stated. "Now, if you go all the way down, we have a few clients now where, because we’re so fast, we can be directly, basically in the process, like in in the processor on the GPU right, and the GPU calculate some output, and it can immediately request from us in a distributed fashion. ‘Oh, I need, I need, I need a human to look at this.’ I get the answer and then apply that loss, which has not been possible so far." Currently, the Rapidata platform facilitates live feedback from approximately 5,500 humans per minute, directly influencing models running on thousands of GPUs. This real-time feedback mechanism is crucial for preventing "reward model hacking," a phenomenon where two AI models might deceptively learn to trick each other within a feedback loop. By grounding the training process in actual, immediate human nuance, Rapidata ensures the integrity of the learning process.
As AI technology progresses beyond straightforward tasks like object recognition and delves into the realm of generative media, the demands placed on data labeling have evolved significantly. The focus has shifted from objective tagging to subjective, "taste-based" curation. The question is no longer simply "Is this a cat?" but rather more nuanced inquiries such as "Is this voice synthesis convincing?" or "Which of these two summaries feels more professional?" Lily Clifford, CEO of the voice AI startup Rime, has attested to the transformative impact of Rapidata in testing models within realistic, real-world contexts. "Previously, gathering meaningful feedback meant cobbling together vendors and surveys, segment by segment, or country by country, which didn’t scale," Clifford remarked. By leveraging Rapidata, Rime can now effectively reach the appropriate audiences – whether they are located in Sweden, Serbia, or the United States – and observe how their AI models perform in actual customer workflows within a matter of days, not months. Corkill further elaborated on this point, noting, "Most models are factually correct, but I’m sure you’ve received emails that feel, you know, not authentic, right? You can smell an AI email, you can smell an AI image or a video, it’s immediately clear to you… these models still don’t feel human, and you need human feedback to do that." This underscores the critical role of human perception in imbuing AI-generated content with a sense of authenticity and naturalness.
From an operational perspective, Rapidata positions itself as a foundational infrastructure layer that liberates companies from the burden of managing their own bespoke annotation operations. By offering a scalable and accessible network, the company is effectively lowering the barrier to entry for AI teams that have historically struggled with the prohibitive costs and inherent complexities associated with traditional feedback loops. Jared Newman of Canaan Partners, who co-led the seed funding round, articulated this strategic importance, suggesting that this type of infrastructure is indispensable for the next generation of AI development. "Every serious AI deployment depends on human judgment somewhere in the lifecycle," Newman stated. "As models move from expertise-based tasks to taste-based curation, the demand for scalable human feedback will grow dramatically."
While the current commercial focus of Rapidata is primarily on AI model laboratories, particularly those concentrated in the Bay Area, Corkill envisions a future where AI models themselves become the principal consumers of human judgment. He terms this concept "human use." In this forward-looking paradigm, an AI designed for car customization, for instance, would not merely generate a generic vehicle design. Instead, it could programmatically engage Rapidata to solicit opinions from a diverse group of 25,000 individuals within a specific market, such as France, regarding a particular aesthetic. The AI would then iteratively refine its design based on this feedback, all within a matter of hours. Corkill astutely observed, "Society is in constant flux. If they simulate a society now, the simulation will be stable for and maybe mirror ours for a few months, but then it completely changes, because society has changed and has developed completely differently." By establishing a distributed and programmatic pathway to access human cognitive capacity on a global scale, Rapidata is strategically positioning itself as the essential bridge between silicon-based intelligence and the dynamic, evolving nature of human society. With its recently secured $8.5 million in seed funding, the company is poised for aggressive expansion, aiming to ensure that as AI continues to scale, the human element is no longer a restrictive bottleneck but rather a real-time, integral feature of its development.