Sapien: Human Powered Axie Infinity For AI - TOKEN2049 Singapore 2024

Summary

Trevor, a web3 enthusiast from Toronto, discusses his journey from early Ethereum days to creating Polymath, an L1 blockchain. He emphasizes the critical role of data labeling in AI, comparing it to oil as a competitive advantage. Trevor unveils Sapien, a decentralized data labeling platform aiming to democratize access to gig work, enabling anyone to earn a living wage by labeling data on their phones. He also highlights the platform's potential to enhance AI training with diverse, high-quality data and introduces innovative web3 features like 'stake to play' to ensure quality and incentivize genuine participation.

Takeaways

• 🌐 Trevor is a web3 veteran, having been involved in the early Ethereum days and co-founded projects like Polymath and Polymesh.
• 🏗️ His latest project is Sapien, a decentralized data-labeling platform that aims to provide a marketplace for data labeling, allowing anyone with a phone to label data and earn money.
• 💻 Trevor believes that in AI, the competitive advantage lies in data, especially human-labeled data, not compute or algorithms.
• 🧠 Human data labeling is critical for improving AI models, and demand for labeled data is growing rapidly, with models spending billions on data labeling annually.
• 📊 Large tech companies use centralized facilities for data labeling, often in Southeast Asia, with workers earning low wages and working under less-than-ideal conditions.
• 💡 Sapien’s decentralized approach allows for a more diverse, global pool of labelers, which leads to higher-quality data and helps close the performance gap between large and smaller AI models.
• 📱 Sapien aims to make data labeling accessible and profitable for everyone, from low-wage workers to highly skilled professionals, enabling a new form of gig work.
• 🌍 Diversity in data labeling is key to higher-quality results, as centralized models tend to have homogeneous data labelers, leading to biased and suboptimal data.
• 🔗 Sapien integrates with web3 ecosystems, offering users the ability to get paid instantly on-chain and utilizing decentralized finance mechanisms like staking to ensure data quality.
• 📈 Trevor emphasizes that human knowledge and reasoning are crucial for advancing AI towards AGI (Artificial General Intelligence), as current open web data is insufficient for training AI in complex reasoning.

Q & A

What was Trevor's involvement in the early Ethereum days?

Trevor was part of the early Ethereum community when Vitalik Buterin was distributing the Ethereum white paper around 2014.

What was Trevor's last web3 project before moving into AI?

Trevor's last web3 project was Polymath, which focused on security tokens (now known as real-world asset tokens). Polymath later evolved into an L1 blockchain called Polymesh, which Trevor co-created with Charles Hoskinson.

Why does Trevor believe data is a sustainable competitive advantage in AI?

Trevor argues that while compute power and algorithms can be easily replicated or bought, data—especially proprietary and well-labeled data—is a sustainable competitive advantage because it is critical for training AI models.

What role does Trevor see for humans in the AI industry?

Trevor emphasizes that humans are crucial for AI, especially in data labeling. He explains that large AI models require human input to improve accuracy and reasoning, contradicting the idea that machines will replace human input.

Why does Trevor believe the internet's current data is insufficient for AGI?

Trevor believes that most internet data shows the result of human reasoning rather than the process of reasoning itself. He suggests that this limitation is one reason AI has not yet reached Artificial General Intelligence (AGI).

What is Sapen, and what problem does it solve?

Sapen is a decentralized marketplace for data labeling. It connects enterprises that need labeled data with people who can label data, making it easier and more accessible for smaller models to compete with big models.

How does Sapen aim to improve the quality of data labeling?

Sapen aims to improve data labeling quality by decentralizing the workforce and ensuring diversity among labelers. More diverse labelers contribute to higher-quality, less-biased data.

What is the potential impact of Sapen’s model on global workforces?

Sapen’s model allows people anywhere in the world to earn a living wage by labeling data, with no specific skills or assets required. It offers a new form of gig work where people can earn money from their phones, democratizing access to income.

What is Trevor’s vision for the future of AI data labeling?

Trevor envisions the future of data labeling as a gamified, decentralized process where individuals can monetize their unique knowledge and skills to improve AI. He sees this as the next evolution of gig work and a key component in advancing AI.

What are some of the specific use cases Trevor has seen for Sapen?

Sapen has been involved in diverse use cases, including labeling data for a fashion company to predict trends, labeling EtherScan data for a DeFi project, and working with a pancreatic cancer researcher. This variety demonstrates the wide applicability of data labeling in different industries.