USC Professor’s Crypto Startup ‘Sahara’ Raises $6M to Reward AI Trainers

• Sahara, describing itself as a “decentralized AI network,” has raised $6 million in a seed funding round led by Polygon Capital.

• The Los Angeles-based startup was co-founded by Sean Ren, an AI researcher and tenured member of the school’s computer science faculty at the University of Southern California (USC).

Sahara, the latest startup to blend the worlds of crypto and artificial intelligence, says it can help workers and companies get compensated for their knowledge, expertise and data in the age of AI.

The Los Angeles-based startup has raised $6 million in a seed funding round led by Polychain Capital. The round also included participation from Samsung Next, Matrix Partners, Motherson Group and Sandeep Nailwal, co-founder of the Polygon blockchain ecosystem.

AI products like ChatGPT and Google’s Gemini are trained on massive troves of data, and they improve over time as a result of user feedback. According to Sahara co-founder Sean Ren, his startup is building a suite of products that will directly reward the users, data providers and AI trainers that help the tools develop – a stark difference from the current model, whereby an AI model’s proceeds are generally directed towards whatever company created it.

“Users provide very personalized feedback to AI systems” to help them improve, explained Ren, who started Sahara after spending seven years as an AI researcher at the University of Southern California (USC), where he is a tenured member of the school’s computer science faculty. People tend to be “totally fine with sharing that data” since they’re just happy to use a free service, Ren told CoinDesk in an interview. But recently, ”we’ve seen people’s jobs start getting replaced due to the fact that they have improved AI by so much.”

Ren says he doesn’t think today’s AI giants are incentivized to compensate their users. As far as finding a solution to the growing problem of people losing their livelihoods to AI, “I think it has to come from the bottom, with some disruptive technology,” Ren said.

According to Ren’s bio on the USC website, he spends time at Allen Institute for AI – founded by the philanthropist and Microsoft co-founder Paul Allen – working on machine common sense, and previously was a data science advisor at Snapchat. His academic research “seeks to build generalizable natural language processing (NLP) systems that can handle a wide variety of language tasks and situations, to broaden the scope of model generality,” the bio reads.

The startup’s other co-founder, Tyler Zhou, studied at the University of California, Berkeley, and worked as an investment director at Binance Labs in 2022 and 2023, according to his LinkedIn profile.

AI model training

One of Sahara’s first products, Sahara Knowledge Agent (KA), will be a customizable AI agent for individuals and businesses. The key difference with Sahara KA relative to other AI agents will be how it’s monetized: The experts and companies that train a Sahara KA can be directly compensated for training the model.

The feature will purportedly use blockchain networks as a way to track data and reward contributors in a privacy-preserving way. It will be built using Sahara Data, another early product from the startup which “provides high-value data services for AI model training, addressing concerns about security and privacy in data handling,“ the company explained in a statement.

According to Sahara, its tech has already seen adoption from the likes of Microsoft, MIT and USC, and two enterprise clients have already begun using the Sahara KA service. Ren says Sahara’s first products will begin rolling out to the wider public over the second and third quarters of this year.

“I think the economy for the Web2 AI industry is kind of broken, and we are trying to make sure every party has their own role in this new ecosystem of Sahara,” said Ren.

“Professional experts, they can really scale up their knowledge – monetize their knowledge – using the agent technology, for example, to reach to more people,” he said. “For a large majority of the, say, laymen who don’t have super strong professional knowledge or expertise, they can contribute to what we call the public knowledge base or the public base models,” which are more general AI agents, he added.

If the model makes money…

In each case, contributors get “a share of the model,” said Ren. “If the model makes money in the future” as a result of people paying to use it, then the “shareholders will get profit.”

Sahara’s seed raise comes as the intersection between AI and crypto is having a moment.

Worldcoin, the crypto startup from Open AI founder Sam Altman, has consistently nabbed headlines with its controversial, retina-scanning approach to verifying digital identities in the age of AI. Vitalik Buterin, the Ethereum co-founder and blockchain thought leader, has been widely cited for a recent blog post and other public statements exploring how AI and crypto can complement one another.

The sector has also seen interest in dollar terms. In addition to several recent fundraises for AI-linked crypto firms, AI-related crypto tokens have surged lately. Notably, they outpaced the CoinDesk 20 index, which benchmarks the biggest cryptocurrencies, on the news last month of strong quarterly earnings for NVIDIA (NVDA), the leading AI chipmaker.