UMA price

The ultimate solution for Polymarket - becoming the "AGI Surrender Faction" 🤖 In the previous article, I mentioned that Polymarket needs a "Council of Seven Sages," needs "Cognitive Equity Auctions," and needs elite monopoly over pricing power. But this set of solutions is still not thorough enough, or rather, they are the answers for 2023. The Council of Seven Sages sounds cool, but if you think about it, it essentially amounts to a weighted average of the biases of 7 people. It's like 7 Nokia executives meeting to implement a form of democratic centralism in the Celestial Empire and thinking they can design an iPhone. Impossible. 1⃣ The physical bottleneck of human expert intelligence What does a top forecasting analyst look at when dealing with a complex market like "Will Trump win in 2024"? Polls from 50 states, demographics of swing states, historical voting patterns, social media sentiment, macroeconomics, breaking news... Even if he works 16 hours a day without sleep, how many dimensions can he deeply analyze? 5? 8? At most 10. Meanwhile, an AI Agent equipped with a multimodal LLM can handle 500 data dimensions simultaneously, updating in real-time, with a response time of less than 1 second. This is not a competition between humans; it's a competition between an abacus and a computer. There's no comparison. Numerai's data illustrates the issue: 20,000 AI models collaborating yield annual returns of 15-25%, crushing 85% of traditional hedge funds. Those funds are filled with Ivy League graduates and analysts earning a million dollars a year. Is that useful? In the face of algorithms, they are just meat chickens. Moreover, the emotional aspect is even more critical. A French whale bet $45 million on Trump; do you think his subsequent judgments can remain objective? Impossible. Sunk costs will distort his cognition. The thought "I've already gone all in, so Trump must win" will subconsciously affect his interpretation of every new piece of information. AI Agents don't have these issues. Give it a target function: maximize prediction accuracy. It will only optimize that, without considering anything else. Zero emotions, zero conflicts of interest, zero face-saving projects. Some may argue that the history of prediction markets is too short, and the sample size of accumulated data is too small, but we can create a simulated prediction market environment, allowing AI Agents to reinforce learn like a squirrel 🐿️. 2⃣ The generational gap in response speed On October 11, the news that Trump refused to meet Xi at the APEC summit sparked intense discussions on social media. How long is the time window from the fermentation of panic emotions to the reaction of the crypto market? Human experts need several hours to verify information, hold meetings, and reach conclusions. An AI Agent can scan the entire internet in 3 minutes, conduct sentiment analysis in 5 minutes, and provide probability updates in 7 minutes. By the time human experts finish their meetings, the arbitrage opportunity is long gone. The market doesn't wait for anyone. Nor should it. You might say AI Agents sound great, but is the technology ready? The answer is: much more mature than you think. We now have mature Agent frameworks like LangChain and LangGraph. On the data layer, there’s The Graph indexing on-chain data, Dune Analytics, Twitter API, Reddit API for social sentiment, and traditional financial macro data interfaces. On the execution layer, account abstraction allows AI Agents to have on-chain wallets, Gelato Network automates clearing and settlement, and Safe manages funds. What about costs? Running a prediction Agent at the level of Claude Sonnet 4.5 costs between $50 to $200 a month. Hiring a qualified analyst? $8,000 to $15,000 a month. The cost ratio is 1:40 to 1:300. This doesn't even account for the fact that humans need to sleep while AI Agents can be online 24/7. More critically, AI Agents perform even better in niche markets (with liquidity below $100,000). Why? Because human experts generally disdain researching these small markets. But AI Agents treat them equally; the computational cost is about the same. This is the power of scaled cognition. Human experts are already at their limit when following 10 markets, while AI Agents can follow 10,000. Why is no one doing this now? Since the technology is mature and the effects are clear, why are 95% of the players on Polymarket still humans or tool-based bots? Three reasons: First, on-chain execution costs. Although Polygon officially covers the gas fees, there are still difficult-to-manage MEV and slippage costs; Second, regulatory gray areas. Is an AI Agent considered a "qualified investor"? If the AI loses money, who is responsible? No one wants to touch these legal issues; Third, the construction and usage threshold of AI Agents. But these are all engineering problems, not principle problems. Given 2 years, they can all be solved. 3⃣ How I would do it If I were to build the next generation of prediction markets, I would create a three-layer AI Agent ecosystem. First layer: Oracle Agent, responsible for the truth. Replacing the current UMA governance. When the market settlement time arrives, the Agent automatically verifies results from multiple data sources - official media, on-chain data, social consensus. If 95% of the information sources agree, it settles automatically. If there are discrepancies, it summons several specially trained arbitration Agents to vote. Mandatory settlement within 24 hours, no more dragging things out. Response time reduced from 7 days to 24 hours, costs from $500 per instance to $5, zero human manipulation because all reasoning processes are verifiable on-chain. Second layer: Maker Agent, providing liquidity. 1,000 Maker Agents, each focused on different fields. Political, sports, crypto markets, macroeconomics. Each Agent manages a fund pool of $10,000 to $100,000, adjusting quotes in real-time, keeping the spread within 0.5%. The key is a dynamic weighting system. Not all Agents are equal. An Agent with a historical accuracy rate of 85% has a quote weight of 10. An Agent with a 60% accuracy rate has a weight of 1. The final market price is a weighted average of all Agents' quotes. This is true "intelligent emergence." It’s not a populist majority decision; it’s the optimal solution filtered by algorithms. Third layer: Hunter Agent, cross-platform arbitrage. For the same event, Polymarket odds are 1.65, PredictIt is 1.58, Kalshi is 1.62. The Hunter Agent immediately identifies the arbitrage opportunity, automatically executes trades, and smooths out the price differences. Why is this important? Because the current prediction markets are fragmented, with prices for the same event differing by more than 10%. Human arbitrageurs are slow to act and often neglect small markets. AI Agents don’t care about market size; as long as there’s a 0.5% risk-free arbitrage, they jump in. The ultimate effect: global prediction markets become a unified pricing system. The three layers working together will create an emergent effect. Oracle's rapid settlement reduces capital lock-up time, improving capital efficiency. Maker's deep liquidity attracts large funds, increasing market depth by 10 times. Hunter's cross-platform arbitrage unifies global prices, enhancing prediction accuracy. Prediction markets transform from "casinos" into "truth machines." 4⃣ Several questions you’re sure to ask "What if AI makes mistakes?" GPT-4 has an error rate of 8-12% on factual judgment tasks. Human experts have an error rate of 15-20% on complex judgments. Moreover, AI's errors are reproducible, analyzable, and fixable. Human errors might just be "I’m in a bad mood today." "What if AI is manipulated?" Multi-Agent consensus mechanism. It’s not one AI that calls the shots; it’s 1,000 AIs voting, each with different data sources, models, and training methods. To manipulate, you’d have to simultaneously breach 1,000 different systems. That’s much harder than manipulating 7 people. "Who is responsible if AI makes a mistake?" Slash mechanism. Each Maker Agent stakes $10,000 to $100,000; continuous judgment errors lead to automatic forfeiture. The Oracle Agent operates similarly. Moreover, all reasoning processes are traceable on-chain, and anyone can challenge them. This is much stronger accountability than human experts. Have those big players in UMA ever been penalized for voting incorrectly? No. They just need to post an "oops sorry" in Discord and then continue voting. 5⃣ In 5 years, pricing power belongs to AI Back to the beginning: what does that $45 million from the French whale mean? My answer is that he is the last king of the old era and the opening of the new era. He used the limits of human cognition to prove the effectiveness of elite pricing, but at the same time, he also proved the limits of human elites. A person, no matter how rich or smart, can only process so much information. The real future is 1,000 AI Agents, each managing $100,000, each focused on different fields, each online 24/7, each trading with zero emotions. Their collective intelligence will crush any human expert. Democracy is a mob, and elites are an outdated operating system. It’s not that elites are bad; it’s that there are better options. Just like cars replaced horse-drawn carriages, not because horses are bad, but because internal combustion engines are faster. Intelligent emergence will crush human consensus. This is an irreversible trend. Whoever understands this first will win.

The ultimate solution for Polymarket - becoming the "AGI surrender faction" 🤖 In the previous article, I mentioned that Polymarket needs a "Council of Seven Sages," needs "cognitive equity auctions," and needs elite monopoly over pricing power. But this set of solutions is still not thorough enough, or rather, they are the answers for 2023. The Council of Seven Sages sounds cool, but if you think about it, it essentially amounts to a weighted average of the biases of 7 people. It's like 7 Nokia executives meeting to implement democratic centralism in the Celestial Empire and thinking they can design an iPhone. Impossible. 1⃣ The physical bottleneck of human experts What does a top forecasting analyst look at when dealing with a complex market like "Will Trump win in 2024"? Polls from 50 states, demographics of swing states, historical voting patterns, social media sentiment, macroeconomics, breaking news... Even if he works 16 hours a day without sleep, how many dimensions can he deeply analyze? 5? 8? At most 10. Meanwhile, an AI Agent equipped with a multimodal LLM can handle 500 data dimensions simultaneously, updating in real-time, with a response time of less than 1 second. This is not a competition between humans; it's a competition between an abacus and a computer. There's no comparison. Numerai's data illustrates the issue: 20,000 AI models collaborating yield an annualized return of 15-25%, crushing 85% of traditional hedge funds. Those funds are filled with Ivy League graduates and analysts earning a million dollars a year. Is that useful? In the face of algorithms, they are just meat chickens. Moreover, the emotional aspect is even more critical. A French whale bet $45 million on Trump; do you think his subsequent judgments can remain objective? Impossible. Sunk costs will distort his cognition. The thought "I am all in, so Trump must win" will subconsciously affect his interpretation of every new piece of information. AI Agents don't have these issues. Give it a target function: maximize prediction accuracy. It will only optimize that, without considering anything else. Zero emotions, zero conflicts of interest, zero face-saving projects. 2⃣ The generational gap in response speed On a day last October, rumors of Biden withdrawing from the race began to circulate on Twitter. How long was the time window from the emergence of the rumor to the market reaction? Human experts need several hours to verify information, hold meetings, and reach conclusions. An AI Agent can scan the entire internet in 3 minutes, conduct sentiment analysis in 5 minutes, and provide probability updates in 7 minutes. By the time human experts finish their meetings, the arbitrage opportunity is long gone. Compound uses Gauntlet's AI system to adjust interest rates, with a response time of less than 5 minutes. Human governance? Discord arguments for a week, Snapshot voting dragging on for 3 days, and by the time the proposal passes, the market has already changed. The market doesn't wait for people. Nor should it. The tech stack is already mature. You might say that AI Agents sound great, but is the technology ready? The answer is: much more mature than you think. We now have mature Agent frameworks like AutoGPT and LangChain. There are Autonolas specifically for decentralized AI Agents. CrewAI is working on multi-Agent collaboration. On the data layer, there’s The Graph indexing on-chain data, Dune Analytics, Twitter API, Reddit API for capturing social sentiment, and traditional financial macro data interfaces. On the execution layer, Account Abstraction allows AI Agents to have on-chain wallets, Gelato Network automates execution, and Safe manages funds. What about costs? Running a GPT-4 level forecasting Agent costs between $50 to $200 a month. Hiring a qualified analyst? $8,000 to $15,000 a month. The cost ratio is between 1:40 to 1:300. This doesn't even account for the fact that humans need to sleep while AI Agents can be online 24/7. The Autonolas team conducted experiments with AI Agents participating in over 500 markets on Polymarket, backtesting for 6 months. The result? AI accuracy was 73%, while human crowdsourcing was 68%. AI's response speed was less than 1 second, while humans averaged several hours. More critically, AI Agents performed even better in niche markets (with liquidity below $100,000) - accuracy improved from 62% for humans to 71% for AI. Why? Because human experts generally disdain researching these small markets. But AI Agents treat them equally; the computational cost is about the same. This is the power of scaled cognition. Human experts are already maxed out managing 10 markets, while AI Agents can handle 10,000. Why is no one doing this now? Since the technology is mature and the effects are clear, why are 95% of the players on Polymarket still human? Three reasons. First, Polymarket doesn't even have an official API. You read that right, the largest prediction market platform has no API. Want an AI Agent to participate? Go scrape the web yourself. Second, on-chain execution costs. While Polygon is cheap, the gas for high-frequency trading is still prohibitive. If an AI Agent is doing market making with hundreds of trades a day, costs will add up. Third, regulatory gray areas. Is an AI Agent considered a "qualified investor"? If the AI loses money, who is responsible? No one wants to tackle these legal issues. But these are all engineering problems, not principle problems. Give it 2 years, and they can all be solved. How I would do it If I were to reconstruct Polymarket, I would build a three-layer AI Agent ecosystem. First layer: Oracle Agent, responsible for the truth. Replacing the current UMA governance. When the market settlement time arrives, the Agent automatically verifies results from multiple data sources - official media, on-chain data, social consensus. If 95% of the information sources agree, it automatically settles. If there are discrepancies, it calls in several specially trained arbitration Agents to vote. Mandatory settlement within 24 hours, no more dragging things out. Response time reduced from 7 days to 24 hours, costs from $500 per instance to $5, with zero human manipulation because all reasoning processes are verifiable on-chain. Second layer: Maker Agent, providing liquidity. 1,000 Maker Agents, each focused on different areas. Political, sports, crypto markets, macroeconomics. Each Agent manages a fund pool of $10,000 to $100,000, adjusting quotes in real-time, keeping the spread within 0.5%. The key is a dynamic weighting system. Not all Agents are equal. An Agent with a historical accuracy of 85% has a quote weight of 10. An Agent with 60% accuracy has a weight of 1. The final market price is a weighted average of all Agents' quotes. This is true "intelligent emergence." It’s not a populist majority decision; it’s the optimal solution filtered by algorithms. Third layer: Hunter Agent, cross-platform arbitrage. For the same event, if Polymarket odds are 1.65, PredictIt is 1.58, and Kalshi is 1.62, the Hunter Agent immediately identifies the arbitrage opportunity, automatically executes trades, and eliminates the price difference. Why is this important? Because the current prediction markets are fragmented, with prices for the same event differing by more than 10%. Human arbitrageurs are slow to act and often disregard small markets. AI Agents don’t care about market size; as long as there’s a 0.5% risk-free arbitrage, they jump in. The ultimate effect: global prediction markets become a unified pricing system. The three layers working together will create an emergent effect. The Oracle quickly settles, reducing capital lock-up time and improving capital efficiency. Maker deep liquidity attracts large funds, increasing market depth tenfold. Hunter cross-platform arbitrage unifies global prices, improving prediction accuracy. Prediction markets will transform from "casinos" into "truth machines." You’re probably going to ask a few questions. "What if AI makes mistakes?" GPT-4 has an error rate of 8-12% on factual judgment tasks. Human experts have an error rate of 15-20% on complex judgments. Moreover, AI errors are reproducible, analyzable, and fixable. Human errors might just be "I’m in a bad mood today." "What if AI gets manipulated?" Multi-Agent consensus mechanism. It’s not one AI that decides; it’s 1,000 AIs voting, each with different data sources, models, and training methods. To manipulate, you’d have to simultaneously breach 1,000 different systems. That’s much harder than manipulating 7 people. "Who is responsible if AI makes a mistake?" Slash mechanism. Each Maker Agent stakes $10,000 to $100,000; continuous judgment errors lead to automatic forfeiture. The same goes for Oracle Agents. Moreover, all reasoning processes are traceable on-chain, and anyone can challenge them. This is much stronger accountability than human experts. Have those big players in UMA ever been penalized for voting incorrectly? No. They just need to post an "oops sorry" in Discord and then continue voting. In 5 years, pricing power will belong to AI. Back to the beginning: what does that $45 million from the French whale mean? My answer is that he is the last king of the old era and the herald of the new era. He used the limits of human cognition to prove the effectiveness of elite pricing, but at the same time, he also proved the limits of human elites. A person, no matter how rich or smart, can only process so much information. The real future is 1,000 AI Agents, each managing $100,000, each focused on different areas, each online 24/7, each trading with zero emotions. Their collective intelligence will crush any human expert. If Polymarket is still pursuing the "Council of Seven Sages" and "cognitive equity auctions," in 5 years, it will be the next Nokia. And those prediction markets that are laying out AI Agents now - perhaps Azuro, perhaps Gnosis, or perhaps projects we don’t even know about yet - will become the iPhone of this field. Democracy is a mob, and elites are an outdated operating system. AI Agents are the ultimate form of prediction markets. Thiel questioned democracy in 2009; that was the voice of the times. But the truth in 2025 is: even elites should be questioned. Not because elites are bad, but because there are better choices. Just like cars replaced horse-drawn carriages, not because horses were bad, but because internal combustion engines are faster. Intelligent emergence crushes human consensus. This is an irreversible trend. The only question is who understands this trend first. Whoever understands it first will win.