“Everyone, everybody, everybody, is our tagline,” Gupta said. “We have organized all the AI models we can find today.” Yupp’s website encourages developers to contact if they wish to add their language or image models to the options. Currently, it has no transactions with AI model builders and can provide these responses by making API calls.
Whenever someone uses Yupp, they participate in head-on comparisons of the two chatbot models, sometimes rewarded for providing feedback and choosing the winning answer. Basically, it’s a user survey disguised as a fun game. (The website has a lot of Emoji. )
In this case, he believes that for users, the trade-offs on data are clearer than in the past consumer applications, such as Twitter, he quickly told me that he is his 27th employee and now owns Biz Stone, one of the company’s co-founders, who is one of his supporters. “It’s a little different from previous consumer companies,” he said. “You provide feedback data, which will be used anonymously and sent to model builders.”
This brings us to the real money position: selling human feedback to AI companies that are eager to have more data to fine-tune their models.
“Crowdsourcing human assessments are what we do here,” Gupta said. He estimates that the amount of cash users can earn will add up and drink a few cups of coffee a month. However, such data markers are often called reinforced learning from human feedback in the AI industry and are very valuable when companies release iterative models and fine-tune their output. It’s worth a lot more than the wettest coffee in San Francisco.
Yupp’s main competitor is a website called LMARENA, which is popular among AI insiders, which is to get feedback on new models and bragging rights if the new version rises to the top of the packaging. Whenever a powerful model is added to LMARENA, it often raises rumors about which major company is trying to test its new version in stealth.
“It’s a two-way product with network impact that helps consumers of model builders,” Gupta said. “Hope builders want to improve the model and submit it to consumers.” He showed me the beta version of Yupp’s rankings, which came online today, which includes the overall ranking of the model and more granular data. Ranking can be filtered by the performance of the model’s performance of a specific demographic information shared with a specific demographic information shared during the registration process or in a specific prompt category (such as health care-related issues).
At the end of the conversation, Gupta proposed artificial general intelligence (super intelligence, theory of human-like algorithms), i.e., the looming technology. “These models are built for human users in the near future, at least,” he said. Although many researchers are still questioning whether the underlying technology behind large language models can produce AGI, this is a fairly common belief and a marketing point.
Gupta hopes Yupp’s users may be anxious about the future of humanity, and they envision themselves as actively shaping these algorithms and improving their quality. “It’s better than freedom because you’ve done this great thing for the future of AI,” he said. “Now, some people will want to know this, while others just want the best answer.”
Even more users may just want extra cash and are willing to spend hours giving feedback during chatbot conversations. I mean, $50 is $50.
