Author: 0xJeff, Crypto KOL
Compiled by: Felix, PANews
Prediction has always been a core ability of human evolution—since ancient times, humans have relied on their senses and instincts to predict threats and opportunities in their environment, including detecting predator activity patterns, chances of prey appearing, and seasonal food supply situations, all of which are crucial for survival.
Since then, this predictive model has gradually evolved into the use and planning of tools (such as predicting crop planting, slaughtering, and meat preservation demands), predicting social cues (intent, emotion, behavior), and developing modern tools such as language, science, mathematics, as well as statistics, computing, machine learning, and artificial intelligence, all of which enhance human predictive capabilities.
Prediction markets have particularly evolved into an economic tool—utilizing human predictive abilities to forecast economic, political, and cultural outcomes. Unlike traditional polls, prediction markets like Polymarket and Kalshi leverage economic incentives to obtain accurate predictions, as participants wager real money.
Polymarket attracted nearly 4 billion dollars in bets in the 2024 U.S. election market, outperforming polls in predicting Trump’s victory, reflecting the economic value of crowdsourced predictions.
The same evolution applies to spot and perpetual contract trading, from the rise of CEX to meet the growing global demand for cryptocurrency, to Hyperliquid's recent disruptive developments that offer self-custody and KYC-free services while providing a CEX-like trading experience.
Prediction is a core ability of human evolution. With the rise of AI/machine learning predictive models, the ability to predict events, asset prices, and volatility is significantly improving.
This will bring humanity into the next stage of evolution.
DeFi 3.0
DeFi 1.0 introduced smart contracts and decentralized applications, allowing anyone to transfer, buy, sell, stake, lend, and yield farm anytime and anywhere. Essentially, it puts crypto assets on-chain to create economic value, such as Uniswap, AAVE, Compound, Curve, Yearn, and Maker.
DeFi 2.0 expanded upon 1.0, introducing novel token economics and incentive distribution mechanisms aimed at coordinating interests among different stakeholders in the protocol (e.g., Olympus/Wonderland, Solidly/Aerodrome) and giving rise to emerging markets that provide alternative yield sources (such as Maple, Pendle, Ethena, Ondo, Clearpool, Solv, USDai, etc.).
DeFi 3.0 introduces artificial intelligence into DeFi. Some call it DeFAI, while others refer to it as AiFi. It means integrating large language models (LLM) and/or machine learning models (ML) into DeFi products.
From simple LLM integration (acting as customer support/co-pilot, helping users navigate protocols) to multi-agent/clusters and machine learning systems, significantly improving products (increasing trading profits, reducing impermanent loss, improving LP yields, lowering liquidation risks for perpetual trades, etc.).
In addition to the DeFAI abstraction layer and fully autonomous financial agents, we will discuss the role of AI/machine learning systems and predictive models in transforming DeFi and other verticals today.
Prediction systems
Neural networks and decision trees have emerged since the 2000s, and these systems were used by hedge funds to predict stock and commodity prices. Early stock prediction results were quite relevant, with short-term prediction accuracy reaching 50% - 60%, but limited data and overfitting restricted their application.
Subsequently, the rise of deep learning and big data has enabled models to handle larger datasets (time series data, unstructured data from news and social media, etc.), leading to more accurate predictions and broader applications.
Breakthrough developments have occurred in the past five years, where Transformer models and multimodal AI have integrated more diverse datasets, such as Twitter sentiment, blockchain transactions, oracles, real-time news, and crowdsourced predictions (Polymarket, Kalshi), leading some AI models to achieve prediction accuracies of 80% - 90% for event outcomes and asset prices.
As these models continue to improve, there is a significant increase in the demand to integrate predictive capabilities into DeFi systems. We are currently in the early stages of DeFi 3.0, witnessing some participants in the market combining AI/machine learning systems with Web3 application scenarios in real time.
DeFi x AI/ML systems
Allora
Allora may be the most widely used decentralized prediction model network at present. Allora has achieved numerous integrations with DeFi protocols and AI agent teams, empowering its predictive capabilities (primarily focusing on cryptocurrency price predictions, such as BTC, ETH, SOL).
Its short-term cryptocurrency price prediction accuracy is reportedly around 80%.
Some major applications include:
Vectis Finance's AI-driven vault based on USDC maximizes SOL trading yields using Allora's reasoning technology. Its cumulative return since April 23 has been 2.4%, with an annual interest rate of about 10%.
Steer Protocol's AI LP vault better positions liquidity ahead of price fluctuations using Allora's predictive price data to avoid impermanent loss.
Allora collaborates with numerous teams, including Cod3x, Axal, Brahma, and Virtuals Protocol, to support trading strategies and execution for AI agents.
Bittensor subnet
Due to Bittensor's dTAO incentive distribution mechanism, which helps startups (subnets) offset development costs, teams utilize Bittensor to kickstart their product development, outsourcing a large amount of development work to miners; the higher the incentives, the better the quality of the miners.
Given that machine learning models and prediction systems are among the easiest tasks to quantify (building models that can accurately predict certain things), this is one of the verticals that subnets focus on the most.
Subnets focused on prediction
SN6 @Playinfgames
SN8 @taoshiio
SN18 @zeussubnet
SN41 @sportstensor
SN44 @webuildscore
SN50 @SynthdataCo
Since SN6, SN18, SN41, and SN44 have been detailed previously, I will skip over these subnets but still want to emphasize:
➔ SN6's @aion5100 (the AI agent/prediction hedge fund layer of SN6) is about to launch a DeFi vault that will automatically allocate user deposits to high-confidence events/markets for betting. This vault is set to launch soon, and the early testing APY is reportedly over four figures.
➔ The @thedkingdao of SN44 shows continuous improvement in signals related to football/soccer. Recent performances in the Club World Cup indicate that aggressive betting has resulted in a 232% return on investment. The team is also working to develop a DeFi vault product that will adopt a more risk-adjusted approach.
The AI agents/tokens representing these two application layers on CreatorBid have excelled in demonstrating the capabilities of SN6 and SN44. This has inspired many other subnet teams to follow suit by launching AI agent tokens to showcase the functionalities of their subnets.
➔ SN50 Synth is particularly interesting. This subnet is built around a highly versatile volatility prediction model. It can cover various probabilities of price occurrences (not just predicting future prices), such as predicting liquidation probabilities, survival times/liquidation times of perpetual positions, setting Univ3 LP ranges and predicting impermanent loss, predicting strike prices and expiration times of options within a window, etc.
Synth is reportedly performing 25% - 30% better than traditional benchmark models (geometric Brownian motion).
There is a huge demand for L1/L2 ecosystems wanting to integrate such engines into their DeFi ecosystems.
So far, Synth has integrated with the following platforms:
Arbitrum supports AI trader competitions
Chainrisk understands volatility so that partner protocols can better cope with significant changes in volatility.
A major liquid staking protocol on Solana for unknown use cases (according to the team, an official announcement will be made in 1-2 days)
The team positions Mode L2 (their own L2) as the application layer, allowing traders to leverage Synth predictions for asset prices and trade better by combining Synth reasoning with Mode AI terminal + Mode Perp products.
SN6, SN44, SN50, and many other subnets are of great interest because they incentivize miners with dTAO tokens worth over 2 million to 10 million dollars annually, encouraging continuous improvement of their prediction models.
Its goal is to use dTAO incentives as capital expenditure to guide product development and achieve commercialization/productization as soon as possible, generating actual revenue and offsetting the selling pressure of dTAO. Some of these subnets have begun to move towards commercialization (as evidenced by DKING providing $300 million in deployment support for a top sports hedge fund).
What will happen next?
The pursuit of higher yields and lower risks will continue, prompting builders to bring more RWAs on-chain. Existing DeFi yield sources will continue to be optimized and become increasingly accessible.
Prediction markets will become a primary source of information, with AI acting as market makers, while experienced participants further stimulate collective intelligence. Tools are becoming increasingly intelligent, and models are becoming more accurate, with some results already seen.
The more these systems learn, the greater their value becomes. Moreover, the stronger their composability with other parts of Web3, the more unstoppable the entire trend becomes.
What this means is... ultimately, everything in the crypto space is a bet on the future.
Thus, infrastructures and applications/agents that can foresee the future even slightly more clearly—whether through collective intelligence, higher quality data, or more accurate models—will have a significant advantage.
Related reading: IOSG: Exploring the competitive landscape of prediction markets through Kalshi
