This Chatbot System Rewards Users with $50 Monthly for Their Input on AI Models

“‘Every AI for everyone’ pretty much sums up our vision,” Gupta explains. “We have compiled every AI model available today.” Yupp’s site invites developers to get in touch if they wish to have their language or image models included. Currently, it has no partnerships with AI model creators, and it delivers responses through API calls.

Whenever someone uses Yupp, they engage in a direct comparison of two chatbot models, sometimes earning rewards for their feedback and selecting a winning response. Essentially, it’s a user survey presented as an enjoyable game. (The website incorporates plenty of emojis.)

Gupta views the trade-off of data in this context as clearer than past consumer applications, such as Twitter—where he notably was the 27th employee and currently counts one of its cofounders, Biz Stone, among his supporters. “This represents a shift from earlier consumer apps,” he says. “You provide feedback data, and that information is anonymized and forwarded to the model creators.”

This leads us to the real revenue source: Selling human feedback to AI firms eager for data to refine their models.

“We’re engaged in crowdsourced human evaluations here,” Gupta details. He believes that the earnings for users could accumulate to cover a few cups of coffee each month. However, this type of data labeling, often referred to as reinforcement learning with human feedback in the AI sector, holds immense value for firms as they release new versions and optimize outputs. It’s worth far more than the fanciest cup of coffee in San Francisco.

Yupp’s main rival is LMArena, a well-liked platform among AI enthusiasts for gathering feedback on new models and gaining bragging rights when a model climbs the ranks. Whenever a robust model is introduced to LMArena, it typically fuels speculation about which significant corporation is secretly testing a new release.

“This is a two-sided platform benefiting both consumers and model creators,” Gupta states. “Ideally, model builders enhance their models and return them to the consumers.” He shares a beta version of Yupp’s leaderboard, launching today, which features an overall model ranking alongside more detailed data. The rankings can be filtered based on user demographics collected during registration, such as age or specific prompt categories like healthcare-related queries.

Toward the end of our discussion, Gupta mentions artificial general intelligence—the concept of superintelligent, human-like algorithms—as a near-future technology. “These models are ultimately being tailored for human users, at least for now,” he remarks. This attitude is common among professionals in AI, although many researchers still debate whether the foundational technology behind large language models can ever achieve AGI.

Gupta hopes that Yupp users, who might be concerned about humanity’s future, will see themselves as actively influencing these algorithms and enhancing their quality. “It’s more than free, because you’re contributing to AI’s future,” he emphasizes. “Some people will appreciate that, while others just want the best responses.”

Furthermore, numerous users may simply be looking for extra cash and willing to dedicate a few hours offering feedback during their chatbot interactions. After all, $50 is $50.