Subnet 06
Infinite Games
Infinite Games
The Infinite Games subnet creates a decentralized forecasting system by incentivizing predictions using advanced LLMs
SN6 : Infinite Games
| Subnet | Description | Category | Company |
|---|---|---|---|
| SN6 : Infinite Games | Future event prediction | Predictive systems | Infinite Games |
The Infinite Games subnet incentivizes the prediction of future events by leveraging recent advances in prediction-focused large language models (LLMs). It aims to develop the first decentralized and self-improving automated forecasting system.
Validators continuously send potential future events to miners, who respond with their predictions. By aggregating real-world synthetic data and complementing the information from prediction markets, the subnet strives to produce highly efficient predictions.
Subnet 6 currently restricts its prediction space to binary future events listed on platforms like Polymarket and Azuro. However, the Giga team has ambitious plans to expand to new markets and providers soon. The primary value of Subnet 6 lies in its ability to enhance the efficiency of prediction markets. Validators can extract this value through arbitrage by obtaining a better understanding of the probability of an event settling and communicating this information to a prediction market by opening a position.
The first applications built on top of Subnet 6 are expected to be closely related to prediction markets. For instance, a trader could query the market to obtain the most up-to-date and relevant predictions for their portfolio based on the current news landscape.
Large Language Models (LLMs) are approaching or even surpassing human forecasting abilities. They perform on par with crowd predictions in prediction markets and exceed human capabilities in predicting neuroscience outcomes. LLMs also demonstrate calibrated confidence in their predictions, meaning they are more confident when they are right. With their generalization capabilities and unbounded information processing, LLMs have the potential to either automate the prediction process or complement human forecasters. LLMs will constantly aggregate the latest and most pertinent news articles, allowing traders to readjust their positions accordingly or trade directly based on this information.
Subnet 6 primarily aims to enhance the efficiency of prediction markets. Validators can capitalize on this by using arbitrage to gain a better understanding of the probability of an event’s outcome and communicate this knowledge to the market by taking positions.
- Incentivized Predictions: Subnet 6 incentivizes the prediction of future events, making it a valuable tool for traders and analysts.
- Judgmental Forecasting: By focusing on judgmental forecasting, Subnet 6 leverages the capabilities of LLMs to provide more accurate and relevant predictions.
- Real-Time Updates: LLMs constantly aggregate the latest news articles, ensuring that users have access to the most current information.
- Arbitrage Opportunities: Validators can extract value through arbitrage, enhancing the efficiency of prediction markets.
- Versatile Applications: From economic forecasts to forward-looking tasks, Subnet 6 offers a wide range of applications for various industries.
Miner’s
Miners compete by sending their event probability estimations to validators. For example, a miner might predict the probability of a specific outcome, like “SBF is sentenced to life.” In the context of Polymarket, an event identifier would be a Polymarket condition ID.
“Approaching Human Level Forecasting with Language Models,” outlines a baseline strategy for miners. The authors fine-tuned an LLM to generate predictions on binary events, achieving near-human performance in submitting forecasts and outperforming human forecasters when the LLM could choose to provide a prediction based on its confidence. The performance of LLMs is likely influenced by the amount of data they can access for a given prediction. If Subnet 6 can continually produce new synthetic data, miners could potentially exceed the current state of the art.
Validator’s
Validators record miners’ predictions and score them once events settle. Each event settlement updates the moving average of a miner’s score. Validators enforce a submission cutoff, currently set at 24 hours before Polymarket events and at the start of sporting events on Azuro, to prevent the probability of the true outcome from approaching one as the event nears resolution.
Scoring Rules
Subnet 6 launched with a quadratic scoring rule (Brier score) for predictions. Validators directly use this rule to score miners based on their predicted probabilities. For events where miners do not submit a prediction, they receive a default score of 0.25.
In the future, a sequentially shared quadratic scoring rule will be implemented. This rule scores miners relative to previous submissions, allowing information aggregation and penalizing miners who do not contribute new information. The aggregated score for each miner, updated at agreed times, will be submitted to the blockchain.
Additionally, a custom scoring rule is being designed to independently score miner progress and stability while avoiding a latency game among miners.
Subnet 6 represents a significant advancement in prediction markets and forecasting. By integrating LLMs and real-time data aggregation, it provides a powerful tool for traders, analysts, and other stakeholders interested in the future of prediction markets
TBC