Pond: Empowering On-Chain Gold Mining with AI

Authors: Shiqi & Michael, Inception Capital

Profiting from on-chain transactions ultimately always involves three steps: discovering new assets, filtering out tokens with a high probability of appreciation, and trading for profit. Pond is a Web3 + AI product that helps users discover new assets and filter out Alpha from them.
What is Pond?
Combining AI and Web3, or using LLMs to provide better Web3 products, is currently the mainstream core of the AI + Web3 narrative. However, the reality is that while LLMs like GPT provide excellent results in interaction and content generation, they are still far from being the all-encompassing AGI.
The core of Pond lies in utilizing industry-native on-chain data and employing technologies such as graph neural networks to learn and predict on-chain behaviors. Through the prediction of behaviors, many new business logics have emerged, such as token price prediction, AI-enhanced MEV, and DeFi strategies.

In terms of price prediction, Pond conducts real-time analysis and data statistics on on-chain tokens, which includes basic information such as market capitalization, trading volume, and changes in the number of holders, as well as on-chain sentiment analysis to reflect the emotional value of tokens. Most importantly, Pond provides a metric called Alpha Rate, the calculation method of which has not been officially disclosed, but this data represents the degree of similarity in trends between a certain token and tokens that have experienced significant price increases recently.
In February, Pond detected a surge in the Alpha Rate of a token called $SYNC, which subsequently saw its price increase 20 times within a month. Therefore, when the Alpha Rate of a certain on-chain project suddenly changes, it may indicate that someone on-chain is preparing to accumulate and pump the token.
How does Pond possess such magical abilities?
Currently, Pond appears to be an atypical AI + Web3 project. Unlike most other projects based on LLMs, Pond is based on a large GNN (Graph Neural Network) to perform real-time statistics and predictions on on-chain data. Compared to LLMs, GNNs are better at processing and generating information, while Pond, based on GNN, excels at uncovering relationships between data and extracting valuable information.
Why choose GNN? Unlike large language models like GPT, GNNs are naturally suited for handling the intricate graph-structured data on blockchains, where on-chain data is formed by complex interactions between accounts and contracts, with wallets and contracts as nodes in the network. Training a GNN model based on on-chain data is not simple, as on-chain data dynamically changes over time. Moreover, if we imagine on-chain transactions as a graph made up of points and lines, there are various diverse relationships between different points. Pond captures the temporal features of transactions and the implicit connections between accounts through innovative technologies, significantly enhancing the model's capabilities.
Who is behind Pond?
Pond is backed by a team of top data scientists and machine learning experts. Team members have published over 40 papers in top international journals and conferences such as IEEE, Nature, and ICML, with an average of over 10 years of experience in the field.
Additionally, Pond's founder, Dylan, previously raised $8M for his last project from Galaxy Digital and founders of several prominent L1s, and established a Crypto community supported by the Ethereum Foundation, giving the Pond team strong community resources in the industry.
Why is Pond worth continuous attention?
First, all interactions between users and applications in Web 3 occur on-chain, and the application value of on-chain data is immense. With the application of machine learning, this vast value will gradually be revealed to the market in previously unexplored ways, unlocking new business logic. The application of on-chain data is not limited to prediction markets; as an early project, Pond has already created various new business logics and collaborated with several well-known projects in different fields. For example, besides mining and predicting potential on-chain Alphas, Pond has also implemented various innovative applications around on-chain interactions, such as AI-based DeFi products, monitoring and preventing on-chain anomalies, and discovering potential marketing opportunities on-chain.
On the other hand, Pond has the potential to bring light to this dark forest of on-chain data, bridging the information gap in the market and transforming abstract transactions hidden behind anonymous addresses into readable and understandable data and phenomena.
Finally, now may be the dawn of the next great narrative on-chain, the prediction market, and Pond is a type of prediction market. Why is it said that prediction markets may be the next grand narrative?
When Vitalik discussed the possibilities of Crypto + AI, he mentioned that for a long time, prediction markets have been the holy grail of cognitive technology. How prediction markets relate to AI, and why is Pond important? Currently, the hottest prediction market in the market is primarily Polymarket, which is based on event-driven betting. In this type of prediction market, it is crucial to have a sufficient supply of betting targets that can cater to various user profiles, data analysis tools that assist in incentivizing users to place bets, and ample liquidity to reduce the costs for participants. To enhance these aspects, besides traditional growth methods, one way is to achieve this through AI. An AI with strong analytical capabilities can assist the platform in creating more reasonable betting targets and recommend them to relevant users based on on-chain data. Additionally, combined with on-chain AI Agents, AI itself can also participate in prediction market activities as a counterparty, providing more liquidity to the market. The foundation for establishing all of this is the prediction of asset and market trends. Previously, whether in academia or practical applications, experiments using AI to predict asset prices have been common, with these predictions trained and forecasted based on various machine learning techniques combined with third-party information sources, such as exchange data. However, there have been few models focused on analyzing, training, and predicting based on on-chain data, and Pond is such a product.
Whether to capture the next Alpha in the market or for research purposes related to the potential next significant narrative of prediction markets in blockchain, Pond is worth experiencing and studying. Currently, Pond is conducting early ecological incentive and user invitation programs, and developers can fill out a form to apply to join Pond's ecosystem, gaining early access to the model and exclusive rewards, while users can also join Pond's invitation program to earn point incentives.