Yupp raises $33 million in seed round! Former Twitter technical backbone reconstructs AI model evaluation system.

a16z crypto led the investment, with Google Chief Scientist Jeff Dean and X co-founder Biz Stone participating. How does Yupp use a crowdsourcing model to break through in AI inclusivity and evaluation?

Written by: KarenZ, Foresight News

With the rapid development of AI technology, the emergence and diversification of AI models have dazzled users. How to choose the right AI model for oneself, in addition to traditional benchmarking, how to enable the model to accurately capture real needs, and how to provide tangible incentives for feedback users have become key challenges that the AI industry urgently needs to address.

Yupp was born in this context as an open platform aimed at establishing an open, transparent, and community-driven AI model evaluation platform. As Yupp states, “Compared to any other technological innovation in history, AI relies more on everyone’s participation and contributions to drive evolution.”

Last week (June 13), Yupp.ai announced a $33 million seed funding round led by a16z crypto, attracting the attention of the community with a roster of high-profile investors including Google Chief Scientist Jeff Dean and X co-founder Biz Stone. At the same time, Yupp launched its product, providing users with a window to explore AI, while redefining the evaluation and optimization methods of AI models through community participation and blockchain technology.

Yupp Team and Financing Background

Yupp is backed by Ber Sarai Labs Inc., co-founded by Pankaj Gupta and Gilad Mishne in June 2024, and has been secretly testing for the past 6 months. The two co-founders and the chief scientist of Yupp met on Twitter in 2010 and have a strong background in the AI industry, having worked at companies like Coinbase, Google, and X.

• Pankaj Gupta: Co-founder and CEO of Yupp, holds a Bachelor’s degree in Computer Science and Engineering from the Indian Institute of Technology Delhi and a Ph.D. in Computer Science from Stanford University. He previously served as a Tech Lead, Senior Manager of Personalization and Recommendations, and Senior ML Engineer at Twitter (March 2009 - May 2014), and held positions as Engineering Director and Senior Engineering Director at Google (July 2017 - March 2021). He was also the first employee and Site Lead at Coinbase India, later serving as Vice President of Engineering and Advisor at Coinbase (April 2021 - May 2024).
• Gilad Mishne: Co-founder and AI Director of Yupp, previously worked as a software engineer at Intel (1998-2000), a senior scientist at Yahoo (2010-2017), a senior engineer and search director at Twitter (2010-2015), and a senior engineering manager and head of machine learning at Google’s Lunar XPRIZE (2019-2023).
• Jimmy Lin: Chief Scientist at Yupp, researched question-answering systems and dialogue interfaces while pursuing a PhD in Computer Science at MIT, currently a professor and David R. Cheriton Chair at the David R. Cheriton School of Computer Science at the University of Waterloo. From 2010 to 2012, Jimmy Lin participated in the infrastructure development for data analysis and data science at Twitter.

This month, Yupp announced that its 33 million seed round financing was completed last year. The Yupp capital matrix covers technology, investment, and academia. In addition to the lead investor a16z crypto, Google Chief Scientist Jeff Dean, X co-founder Biz Stone, Pinterest co-founder Evan Sharp, Perplexity CEO Aravind Srinivas, Cred CEO Kunal Shah, four professors from Stanford University (Dan Boneh, Chris Re, Nick McKeown, Balaji Prabhakar), Othman Laraki, Paul Grewal, Gokul Rajaram, and Coinbase Ventures also participated.

What is Yupp? How does it work?

Yupp is positioned as an AI model exploration and evaluation platform that allows users to experience and compare various AI models for free. Its core concept is to conduct model evaluation through crowdsourcing: users submit prompt words, compare responses generated by different AI models, provide the best answers and evaluation feedback, and receive redeemable point incentives. These choices and feedback will be recorded to form data for the later training and evaluation of AI models.

Yupp will also leverage open-access and permissionless technologies such as blockchain, cryptographic primitives and protocols like zero-knowledge proofs and challenge/response mechanisms, as well as privacy-preserving technologies like confidential computing, to build a system with provable trust neutrality, fairness, and robustness.

The operational mechanism of Yupp can be summarized as follows:

1. Model Exploration and Comparison: Yupp brings together over 500 AI models, including ChatGPT, Claude, Gemini, DeepSeek, Grok, Llama, and more. Users can find various AI models on the platform’s chat page and conduct prompt tests to intuitively compare their strengths and weaknesses side by side.

Yupp currently has two pages: a chat page and a leaderboard. The chat page is designed to be simple, featuring a message box, file upload function, model selection function (optional), image upload function, and a toggle between private and public chat modes (default is private).

As shown in the figure below, after asking a question, Yupp will provide two versions of AI responses, and then the user will select the better version.

It is worth mentioning that when asking questions, the model automatically selects by default and may sometimes hide the model name to collect more objective feedback; of course, users can also ask questions randomly. In addition, Yupp’s QuickTake AI feature can also provide brief summary responses.

2. User Feedback: After users select a better answer, they can further provide feedback on the quality preference of the answer by clicking on the tags, or they can freely express their thoughts in writing. This feedback will help users customize future AI answers on Yupp and also allow Yupp to offer the model for free.

3. Feedback Rewards: After providing feedback, users will receive a points scratch card. Points can be used to ask questions or redeem.

4. Evaluation: Users participate in the evaluation process of the model by selecting the best responses and providing feedback. Yupp aims to establish a publicly transparent evaluation system that allows AI developers to obtain valuable training data, while users can receive rewards, jointly promoting the development of AI technology. The Yupp platform features a public leaderboard called “Yupp VIBE Score” (VIBE, Vibe Intelligence Benchmark), which aims to improve model performance using user feedback while ensuring the privacy of user prompts, unless users choose to share.

Yupp will create a leaderboard based on user feedback and response speed among other factors. In the leaderboard, AI models integrated on the platform can be sorted according to filters, with dimensions including VIBE score, confidence interval of probability samples, voting situation, speed, latency, and input/output cost.

Yupp’s AI model evaluation will combine user preference data to segment users and evaluation data in a more granular way, thus providing samples for AI developers. Yupp stated that, leveraging the founding team’s experience in addressing spam and bot issues on Twitter, they developed complex algorithms to filter out low-quality data, ensuring the integrity of rankings. Yupp has also assembled a dedicated trust and safety team and will continue to invest heavily in this area.

Yupp Points System: Balance Rules for Consumption and Redemption

Yupp points are consumed through questions and earned through feedback, and a portion can be redeemed. Yupp indicates that if users responsibly use Yupp, they will always have enough points to ask the AI model questions and can also redeem a portion of their points as a token of appreciation for improving the ecosystem.

Asking questions requires spending points. After initial registration, you can receive 5000 points for free. Additionally, the total cost is the sum of the default fee, PRO model fee, attachment submission fee, and pre-selected image model fee.

• Basic fee: Each prompt costs 50 points by default. If an image is generated, the fee is 100 Yupp points.
• PRO Model: These advanced models are only available on other platforms through paid subscriptions. If users select the PRO model in advance, an additional 50 Yupp points will be required for each question.
• MAX Model: These models are created using the most expensive models. If a user selects the MAX model in advance, each model per prompt will require an additional 300 Yupp points (including a total base fee of 350 points).
• Attachment submission fee: Each attachment costs 25 Yupp points.
• Select image model: Each pre-selected image model requires an additional cost of 100 Yupp points.

In addition, Yupp Q&A is private by default. If users choose to make the Q&A public, they only need to pay half of the usual fee mentioned above.

As mentioned earlier, the feedback model can earn points scratch cards. The author received several points scratch cards ranging from 200 to 500 points.

Yupp stated that users can withdraw points and exchange them for USD, EUR, INR, and more than 20 other currencies, or convert them into stablecoins (based on Base and Solana). Yupp has also established partnerships with payment service providers such as Stripe, Paypal, and Coinbase to meet the diverse needs of users. Every 1000 points can be exchanged for 1 USD. However, the redemption feature is currently unavailable. To circumvent witchcraft or abuse behavior, Yupp has also set up point redemption rules:

• The maximum number of withdrawals per day is 1, and the withdrawal limit is 10 dollars (10,000 points);
• A maximum of 3 withdrawals per week, with a total withdrawal limit of $20 (20,000 points);
• A maximum of 6 withdrawals per month, with a total withdrawal limit of 50 dollars (50,000 points).

In addition, Yupp stipulates that transactions such as purchasing, selling, trading, or transferring Yupp points that violate the terms of service are deemed invalid and may result in immediate deactivation of the account. Abuse may lead to the disabling of product features and even suspension of the account.

How to participate?

The participation process for Yupp is as follows:

• Register with a Google account (registration grants 5000 points, and the official statement indicates that using the “yupp-launch” code before June 20 will earn an additional 2500 points);
• Ask the AI model to choose the better answer between two AI models;
• Select feedback tags or provide written feedback.
• Use the mouse to scratch the scratch card to receive points.
• Decide whether to withdraw on your own.

Summary

As Chris Dixon, co-founder and general partner of a16z crypto, stated, “Yupp’s design transforms human judgment into a sustainable economic resource. As new interactions replace old data, the data becomes ‘stale’, forming a natural positive feedback loop: more usage brings newer evaluations; newer evaluations generate better models; better models attract more usage. All participants—from users to AI model builders—can engage and see the same transparent rules apply to everyone, ensuring a trustworthy and neutral market. No one can hide the leaderboard, and no one can manipulate rewards or outcomes.”

Yupp’s slogan “Every AI for everyone” can be more accurately translated as “Inclusive AI”. Yupp aims to build the “evaluation infrastructure” of the AI era through blockchain technology and a crowdsourcing model: allowing users to receive incentives through feedback, enabling developers to obtain real data, and ultimately promoting the evolution of AI technology towards a more inclusive and trustworthy direction.

Reference: