With the amount of new subnets being added it can be hard to get up to date information across all subnets, so data may be slightly out of date from time to time
With the amount of new subnets being added it can be hard to get up to date information across all subnets, so data may be slightly out of date from time to time
Infinite Games is a specialized Bittensor subnet focused on decentralized forecasting of future events. In essence, it creates an open market where participants (via AI models) predict outcomes of upcoming events – such as election results, economic indicators, or sports contests – and are rewarded for accuracy. The goal is to develop the first automated, self-improving forecasting system powered by collective intelligence. The team describes their mission as “decod[ing] the future, delivering outcomes before they happen”, reflecting a vision of providing reliable predictions ahead of real-world events.
This subnet leverages advances in large language models (LLMs) to perform what is known as judgmental forecasting – i.e. making probability judgments about one-off or cross-domain events based on reasoning, as opposed to pure time-series prediction. Recent research has shown that LLMs can approach or even surpass human forecasters on such tasks, and Infinite Games builds on this insight. Validators in the network continuously propose questions about potential future events (e.g. “Will the quarterly inflation rate fall below 5%?”), and miners (predictor nodes) respond with probability estimates for those outcomes. Once the real event outcome is known, the validators score the predictions and reward the miners in proportion to their forecasting performance. By incentivizing a swarm of AI-powered predictors to compete and learn, Subnet-6 aims to produce calibrated forecasts that improve over time, essentially crowd-sourcing an ever-evolving “oracle” for future events.
Infinite Games is a specialized Bittensor subnet focused on decentralized forecasting of future events. In essence, it creates an open market where participants (via AI models) predict outcomes of upcoming events – such as election results, economic indicators, or sports contests – and are rewarded for accuracy. The goal is to develop the first automated, self-improving forecasting system powered by collective intelligence. The team describes their mission as “decod[ing] the future, delivering outcomes before they happen”, reflecting a vision of providing reliable predictions ahead of real-world events.
This subnet leverages advances in large language models (LLMs) to perform what is known as judgmental forecasting – i.e. making probability judgments about one-off or cross-domain events based on reasoning, as opposed to pure time-series prediction. Recent research has shown that LLMs can approach or even surpass human forecasters on such tasks, and Infinite Games builds on this insight. Validators in the network continuously propose questions about potential future events (e.g. “Will the quarterly inflation rate fall below 5%?”), and miners (predictor nodes) respond with probability estimates for those outcomes. Once the real event outcome is known, the validators score the predictions and reward the miners in proportion to their forecasting performance. By incentivizing a swarm of AI-powered predictors to compete and learn, Subnet-6 aims to produce calibrated forecasts that improve over time, essentially crowd-sourcing an ever-evolving “oracle” for future events.
Infinite Games delivers a full decentralized prediction platform built on the Bittensor protocol. On a technical level, it defines a custom incentive mechanism where the “commodity” is information – accurate predictions of future binary events. Key aspects of the product include:
Incentivized Predictions: The subnet is designed to reward users (miners) for providing probability forecasts on future events. By tying rewards to accuracy (via proper scoring rules), it encourages honest, well-calibrated predictions rather than random guesses. This creates a competitive marketplace for predictive insight.
Judgmental Forecasting via LLMs: Infinite Games focuses on human-like reasoning tasks – predicting discrete events using broad context and intuition – as opposed to purely statistical modeling. Miners are expected to run large language models that synthesize news, data, and domain knowledge into forecasts. In practice, an LLM-based pipeline generates and updates the event questions and probable outcomes, so the system constantly deals with fresh, real-world scenarios (politics, finance, sports, etc.).
Real-Time Updates: The network supports continuous forecasting. Each event’s prediction window is divided into multiple intervals, requiring miners to update their forecasts as new information comes in. LLM “agents” ingest the latest news and developments, ensuring that predictions reflect up-to-date information. This real-time aggregation of knowledge means users (or applications) can query the subnet at any time for the most current outlook on an event.
Arbitrage and Market Efficiency: A core value proposition is that the subnet’s consensus forecast can enhance the efficiency of prediction markets. Validators (who see all miner predictions) can act on this information advantage – for example, by taking positions on external prediction markets or betting platforms if the subnet’s implied probability differs from the public odds. In other words, Subnet-6 can identify mispricings (e.g. if the crowd underestimates an outcome that the AI predictors strongly favor) and arbitrage them. This feedback loop theoretically pushes external markets toward more accurate odds, demonstrating a practical use-case for the subnet’s output.
Versatile Applications: While the first envisioned applications are in crypto prediction markets, the infrastructure is generic for any forward-looking task. In the near term, a trader or analyst could query Infinite Games for a quick risk assessment on events affecting their portfolio (the system might parse the day’s news and output a probability that, say, a certain policy will pass). In the longer term, validators could offer forecasting-as-a-service – e.g. paid queries for custom economic forecasts, product demand predictions, or other domain-specific outlooks. Essentially, any question about the future that can be posed in binary form could be answered by the network’s collective intelligence.
Technical architecture
The subnet operates as a module on the Bittensor blockchain (Subtensor), with custom logic for handling prediction tasks. It restricts its initial prediction space to binary events that have verifiable outcomes. At launch, these were primarily events listed on decentralized prediction platforms like Polymarket (for general events) and Azuro (for sports outcomes) – leveraging those platforms’ questions as a source of truth and settlement data. The Infinite Games codebase includes connectors that fetch upcoming event IDs and deadlines from these providers, feed them into the Bittensor subnet as tasks, and later retrieve the resolved outcomes for scoring. The team is “actively expanding to new data providers” beyond these initial sources, meaning the subnet will incorporate a broader array of forecasting domains over time (e.g. additional prediction markets or real-world data feeds).
On the client side, miner nodes run an AI model (or ensemble of models) that takes an event description and outputs a probability (0.0 to 1.0) that the event will occur. To discourage last-minute guesses with insider info, the subnet enforces a submission cutoff – for example, miners must submit final predictions at least 24 hours before a given event’s resolution time. Predictions are recorded on-chain by validators, and once the event outcome is known, validators score each miner’s predictions. Initially, the project launched with a quadratic Brier score as the scoring rule (a proper scoring rule that rewards accuracy and confidence calibration). Every prediction contributes to the miner’s rolling performance average, so consistent accuracy is required to earn steady rewards. Notably, the team has since implemented an advanced peer-review scoring mechanism: rather than scoring forecasts solely against the ground truth, a miner’s score on an event is adjusted relative to the predictions of peer miners. (Intuitively, this rewards a miner more if they were confident in a correct outcome that others underrated, and penalizes her if she missed something others caught.) This peer-based scoring aims to encourage diversity and bold, truthful forecasting by using the wisdom of the crowd as an additional benchmark.
Deliverables and integrations:
The Infinite Games project provides a suite of tools around the subnet. The core logic (validators and miners) is open-sourced on GitHub, including documentation for setting up a validator or contributing a custom predictor model. The team has a leaderboard website that displays top-performing predictors and their accuracy stats in real time
leaderboard.infinitegam.es, highlighting that over $1 million USD worth of $TAO (the native token) is available in rewards to be claimed by miners each month
leaderboard.infinitegam.es. There is also a community dashboard for exploring active predictions and outcomes, and an official web interface where users can see example questions and the subnet’s latest consensus (for instance, the current probability the inflation rate stays below 5% as predicted by the network). All these pieces – the smart-contract layer, AI model pipeline, and user-facing interfaces – make up the “product” of Subnet-6, which is essentially a decentralized forecasting service powered by incentivized machine learning.
Updates
Since its launch, Infinite Games has shared updates and progress through public channels and has been steadily evolving. Some of the latest announcements and developments include:
Launch Event: The official launch on June 20, 2024 was a major milestone announced to the Bittensor community. The team celebrated the deployment of Subnet-6 on mainnet after months of preparation. They likely shared this news via their Twitter (X) handle and Discord, inviting miners and validators to join. The launch introduced the key features (LLM-driven forecasting on Polymarket/Azuro events) and effectively opened the subnet for anyone to participate. Early adopters could start staking TAO, running validator nodes, or deploying miner models to begin earning from predictions.
Community Engagement: Post-launch, Infinite Games has been proactive in engaging the community. They maintain an active Twitter account (@PlayInfGames) where they share updates, answer questions, and tease new features. For example, they have discussed the introduction of the peer scoring mechanism and why it improves the system, and they have highlighted particularly interesting prediction questions that the subnet is tackling. They also use Twitter/Discord to announce any new competitions – such as the planned “LLM Guesser Competition” where model developers can compete to climb the leaderboard. Additionally, the team provides support via Discord (an official Discord server is listed on their site), where miners can get technical help, and where the community can discuss prediction strategies or request new features. A Telegram channel is also available for updates.
Integrations and Partnerships: In terms of partnerships, the subnet’s reliance on Polymarket and Azuro data suggests a form of collaboration or at least mutual interest. While not a formal partnership, the Infinite Games team has certainly interacted with these platforms’ APIs and possibly their communities. Any time a new data provider is integrated, it’s announced as an update. For instance, if they add support for a new prediction market or start pulling events from a platform like Manifold Markets or centralized sources (like Metaculus or Good Judgment Project datasets), they will publicize it. These integrations not only expand the subnet’s capabilities but also can be seen as partnerships where Infinite Games brings additional analytical power to those ecosystems. We may see more overt collaborations in the future, such as co-hosted tournaments with a prediction market platform or research collaborations with academic groups (given the project’s overlap with AI forecasting research).
Performance Updates: As the network runs, the team shares progress reports on how it’s performing. This includes disclosing metrics like the average accuracy of predictions, the total number of events predicted, and the volume of TAO rewards distributed. For example, the leaderboard site indicates the scale of rewards (over $1M in TAO monthly) and shows an ongoing accuracy score for the subnet. If certain high-profile events were successfully forecast by the subnet, the team will highlight those wins. Conversely, challenging cases might be discussed to iterate on the models. These updates help demonstrate momentum and build trust that the subnet is producing meaningful results.
Network Growth: By early 2025, Infinite Games has grown its participant base. Updates have noted the number of active validators and miners rising as more people become aware of the opportunity to earn TAO by contributing predictions. For instance, after launch the subnet started with a small set of core contributors (possibly the team and close community members), but over the last few months it has onboarded additional independent miners running their own LLMs. The team has likely updated the community on any significant growth milestones (e.g., “We now have 100+ active miners covering dozens of events each week”), showing that the subnet is decentralizing further. They also encourage outside contributions: their materials include calls to “Contribute your model” and “Become a miner”, emphasizing that anyone with a good predictor model can join and compete.
Future Announcements: Looking forward, we can expect announcements around the publication of the aforementioned benchmark results (which will be a big validation moment for the project). The team might also announce when the competition winners (best models) are determined, potentially open-sourcing top-performing model architectures for others to use. Any major upgrade (for example, moving to support multi-outcome events instead of binary, or enabling an EVM module on Bittensor for smart-contract querying of predictions) would also come through in updates. Lastly, as Infinite Games matures, the team might attend or present at conferences (crypto or AI events) – such appearances would be shared through their channels, highlighting the project’s recognition in wider circles.
Infinite Games delivers a full decentralized prediction platform built on the Bittensor protocol. On a technical level, it defines a custom incentive mechanism where the “commodity” is information – accurate predictions of future binary events. Key aspects of the product include:
Incentivized Predictions: The subnet is designed to reward users (miners) for providing probability forecasts on future events. By tying rewards to accuracy (via proper scoring rules), it encourages honest, well-calibrated predictions rather than random guesses. This creates a competitive marketplace for predictive insight.
Judgmental Forecasting via LLMs: Infinite Games focuses on human-like reasoning tasks – predicting discrete events using broad context and intuition – as opposed to purely statistical modeling. Miners are expected to run large language models that synthesize news, data, and domain knowledge into forecasts. In practice, an LLM-based pipeline generates and updates the event questions and probable outcomes, so the system constantly deals with fresh, real-world scenarios (politics, finance, sports, etc.).
Real-Time Updates: The network supports continuous forecasting. Each event’s prediction window is divided into multiple intervals, requiring miners to update their forecasts as new information comes in. LLM “agents” ingest the latest news and developments, ensuring that predictions reflect up-to-date information. This real-time aggregation of knowledge means users (or applications) can query the subnet at any time for the most current outlook on an event.
Arbitrage and Market Efficiency: A core value proposition is that the subnet’s consensus forecast can enhance the efficiency of prediction markets. Validators (who see all miner predictions) can act on this information advantage – for example, by taking positions on external prediction markets or betting platforms if the subnet’s implied probability differs from the public odds. In other words, Subnet-6 can identify mispricings (e.g. if the crowd underestimates an outcome that the AI predictors strongly favor) and arbitrage them. This feedback loop theoretically pushes external markets toward more accurate odds, demonstrating a practical use-case for the subnet’s output.
Versatile Applications: While the first envisioned applications are in crypto prediction markets, the infrastructure is generic for any forward-looking task. In the near term, a trader or analyst could query Infinite Games for a quick risk assessment on events affecting their portfolio (the system might parse the day’s news and output a probability that, say, a certain policy will pass). In the longer term, validators could offer forecasting-as-a-service – e.g. paid queries for custom economic forecasts, product demand predictions, or other domain-specific outlooks. Essentially, any question about the future that can be posed in binary form could be answered by the network’s collective intelligence.
Technical architecture
The subnet operates as a module on the Bittensor blockchain (Subtensor), with custom logic for handling prediction tasks. It restricts its initial prediction space to binary events that have verifiable outcomes. At launch, these were primarily events listed on decentralized prediction platforms like Polymarket (for general events) and Azuro (for sports outcomes) – leveraging those platforms’ questions as a source of truth and settlement data. The Infinite Games codebase includes connectors that fetch upcoming event IDs and deadlines from these providers, feed them into the Bittensor subnet as tasks, and later retrieve the resolved outcomes for scoring. The team is “actively expanding to new data providers” beyond these initial sources, meaning the subnet will incorporate a broader array of forecasting domains over time (e.g. additional prediction markets or real-world data feeds).
On the client side, miner nodes run an AI model (or ensemble of models) that takes an event description and outputs a probability (0.0 to 1.0) that the event will occur. To discourage last-minute guesses with insider info, the subnet enforces a submission cutoff – for example, miners must submit final predictions at least 24 hours before a given event’s resolution time. Predictions are recorded on-chain by validators, and once the event outcome is known, validators score each miner’s predictions. Initially, the project launched with a quadratic Brier score as the scoring rule (a proper scoring rule that rewards accuracy and confidence calibration). Every prediction contributes to the miner’s rolling performance average, so consistent accuracy is required to earn steady rewards. Notably, the team has since implemented an advanced peer-review scoring mechanism: rather than scoring forecasts solely against the ground truth, a miner’s score on an event is adjusted relative to the predictions of peer miners. (Intuitively, this rewards a miner more if they were confident in a correct outcome that others underrated, and penalizes her if she missed something others caught.) This peer-based scoring aims to encourage diversity and bold, truthful forecasting by using the wisdom of the crowd as an additional benchmark.
Deliverables and integrations:
The Infinite Games project provides a suite of tools around the subnet. The core logic (validators and miners) is open-sourced on GitHub, including documentation for setting up a validator or contributing a custom predictor model. The team has a leaderboard website that displays top-performing predictors and their accuracy stats in real time
leaderboard.infinitegam.es, highlighting that over $1 million USD worth of $TAO (the native token) is available in rewards to be claimed by miners each month
leaderboard.infinitegam.es. There is also a community dashboard for exploring active predictions and outcomes, and an official web interface where users can see example questions and the subnet’s latest consensus (for instance, the current probability the inflation rate stays below 5% as predicted by the network). All these pieces – the smart-contract layer, AI model pipeline, and user-facing interfaces – make up the “product” of Subnet-6, which is essentially a decentralized forecasting service powered by incentivized machine learning.
Updates
Since its launch, Infinite Games has shared updates and progress through public channels and has been steadily evolving. Some of the latest announcements and developments include:
Launch Event: The official launch on June 20, 2024 was a major milestone announced to the Bittensor community. The team celebrated the deployment of Subnet-6 on mainnet after months of preparation. They likely shared this news via their Twitter (X) handle and Discord, inviting miners and validators to join. The launch introduced the key features (LLM-driven forecasting on Polymarket/Azuro events) and effectively opened the subnet for anyone to participate. Early adopters could start staking TAO, running validator nodes, or deploying miner models to begin earning from predictions.
Community Engagement: Post-launch, Infinite Games has been proactive in engaging the community. They maintain an active Twitter account (@PlayInfGames) where they share updates, answer questions, and tease new features. For example, they have discussed the introduction of the peer scoring mechanism and why it improves the system, and they have highlighted particularly interesting prediction questions that the subnet is tackling. They also use Twitter/Discord to announce any new competitions – such as the planned “LLM Guesser Competition” where model developers can compete to climb the leaderboard. Additionally, the team provides support via Discord (an official Discord server is listed on their site), where miners can get technical help, and where the community can discuss prediction strategies or request new features. A Telegram channel is also available for updates.
Integrations and Partnerships: In terms of partnerships, the subnet’s reliance on Polymarket and Azuro data suggests a form of collaboration or at least mutual interest. While not a formal partnership, the Infinite Games team has certainly interacted with these platforms’ APIs and possibly their communities. Any time a new data provider is integrated, it’s announced as an update. For instance, if they add support for a new prediction market or start pulling events from a platform like Manifold Markets or centralized sources (like Metaculus or Good Judgment Project datasets), they will publicize it. These integrations not only expand the subnet’s capabilities but also can be seen as partnerships where Infinite Games brings additional analytical power to those ecosystems. We may see more overt collaborations in the future, such as co-hosted tournaments with a prediction market platform or research collaborations with academic groups (given the project’s overlap with AI forecasting research).
Performance Updates: As the network runs, the team shares progress reports on how it’s performing. This includes disclosing metrics like the average accuracy of predictions, the total number of events predicted, and the volume of TAO rewards distributed. For example, the leaderboard site indicates the scale of rewards (over $1M in TAO monthly) and shows an ongoing accuracy score for the subnet. If certain high-profile events were successfully forecast by the subnet, the team will highlight those wins. Conversely, challenging cases might be discussed to iterate on the models. These updates help demonstrate momentum and build trust that the subnet is producing meaningful results.
Network Growth: By early 2025, Infinite Games has grown its participant base. Updates have noted the number of active validators and miners rising as more people become aware of the opportunity to earn TAO by contributing predictions. For instance, after launch the subnet started with a small set of core contributors (possibly the team and close community members), but over the last few months it has onboarded additional independent miners running their own LLMs. The team has likely updated the community on any significant growth milestones (e.g., “We now have 100+ active miners covering dozens of events each week”), showing that the subnet is decentralizing further. They also encourage outside contributions: their materials include calls to “Contribute your model” and “Become a miner”, emphasizing that anyone with a good predictor model can join and compete.
Future Announcements: Looking forward, we can expect announcements around the publication of the aforementioned benchmark results (which will be a big validation moment for the project). The team might also announce when the competition winners (best models) are determined, potentially open-sourcing top-performing model architectures for others to use. Any major upgrade (for example, moving to support multi-outcome events instead of binary, or enabling an EVM module on Bittensor for smart-contract querying of predictions) would also come through in updates. Lastly, as Infinite Games matures, the team might attend or present at conferences (crypto or AI events) – such appearances would be shared through their channels, highlighting the project’s recognition in wider circles.
Infinite Games was created by an independent team known as Giga (short for Gigaverse). This team is not part of the core Opentensor Foundation, but rather an external builder leveraging the Bittensor network. In the Bittensor subnet directory, Subnet-6 is listed as “Infinite Games – Team: Infinite Games (Giga)”, indicating the project and team share the same name. The Gigaverse group describes itself as providing “ideas and execution for headless orgs,” hinting at a focus on decentralized, community-run technology ventures. They maintain a website and social presence under the Giga brand, and the Infinite Games initiative is one of their flagship projects (possibly alongside other experiments in the Web3/AI space). As of now, detailed information about individual team members is limited.
The Gigaverse (GIGA) collective itself appears to be a newly formed entity, possibly started around 2023, aiming to build “headless” (decentralized) organizations and products. Their “Powered by Giga” footer on the project site and a minimalist Gigaverse webpage hint that they are positioning to incubate multiple projects in the crypto AI realm. Overall, Infinite Games is led by an up-and-coming team of AI and blockchain enthusiasts, operating under the Giga banner, who have chosen Bittensor as the platform to bring their forecasting network to life.
Infinite Games was created by an independent team known as Giga (short for Gigaverse). This team is not part of the core Opentensor Foundation, but rather an external builder leveraging the Bittensor network. In the Bittensor subnet directory, Subnet-6 is listed as “Infinite Games – Team: Infinite Games (Giga)”, indicating the project and team share the same name. The Gigaverse group describes itself as providing “ideas and execution for headless orgs,” hinting at a focus on decentralized, community-run technology ventures. They maintain a website and social presence under the Giga brand, and the Infinite Games initiative is one of their flagship projects (possibly alongside other experiments in the Web3/AI space). As of now, detailed information about individual team members is limited.
The Gigaverse (GIGA) collective itself appears to be a newly formed entity, possibly started around 2023, aiming to build “headless” (decentralized) organizations and products. Their “Powered by Giga” footer on the project site and a minimalist Gigaverse webpage hint that they are positioning to incubate multiple projects in the crypto AI realm. Overall, Infinite Games is led by an up-and-coming team of AI and blockchain enthusiasts, operating under the Giga banner, who have chosen Bittensor as the platform to bring their forecasting network to life.
The Infinite Games subnet had a clear launch and has articulated plans for future development. Here is an outline of its roadmap and anticipated milestones:
Mainnet Launch (Q2 2024): After an extensive period of testnet experimentation and refinement, Infinite Games officially launched on the Bittensor main network on June 20, 2024. This launch introduced the core prediction market mechanics (with Polymarket and Azuro event integration) and the initial Brier scoring mechanism for miner rewards. The subnet went live with a set of active validators and miners, effectively bootstrapping the world’s first decentralized LLM-driven forecasting system. (Prior to mainnet, the team ran closed tests and perhaps community test competitions to tune the parameters and ensure stability.)
Short-Term Improvements (Q3–Q4 2024): In the months following launch, the team focused on broadening the data sources and improving the quality of predictions. The roadmap included integrating a wider array of event streams: beyond Polymarket and Azuro, this could involve other on-chain prediction markets or oracle feeds (for example, real-world event data, additional betting platforms, or even news APIs). The goal is to maximize cross-domain expertise by feeding the predictor models more diverse information. During this phase, the team also planned to enhance the “guessers” (predictor models) themselves. One initiative mentioned is a community competition to attract better LLM-based predictors. By launching an open competition or bounty program, Infinite Games intends to encourage data scientists and AI enthusiasts to fine-tune models that can beat the current ones, thus continuously improving the subnet’s forecasting accuracy. Alongside this, there is work on developing a more user-friendly interface and comprehensive documentation – lowering the barrier to entry for new miners/validators and making the system easier to understand for outsiders.
Scoring Mechanism Upgrade (late 2024): A significant technical update in the roadmap was the introduction of the peer-based scoring system to replace (or augment) the initial Brier score approach. This was implemented to address some edge cases and to further incentivize useful contributions. By late 2024, Infinite Games transitioned to using peer comparison for scoring forecasts, which was an important milestone in optimizing the reward scheme. This change likely came after observing the subnet in action for a few months and identifying ways to better reward true insight (for example, rewarding contrarian but correct predictions more strongly). Such an update demonstrates the “self-improving” aspect of the subnet – not only are the models learning, but the subnet’s rules themselves are being refined over time.
Benchmarking and Results (early 2025): On the roadmap, the team has indicated an intention to publish performance benchmarks for Infinite Games, comparing its forecasting accuracy against industry standards and other forecasting methods. This “Big Benchmark” would be the first comprehensive evaluation of how well the decentralized network of LLM predictors performs relative to, say, traditional prediction markets, expert human forecasts, or academic models. Releasing these results would validate (or inform adjustments to) the approach. We can anticipate a report or paper from the team detailing metrics like Brier score averages, calibration of probabilities, and perhaps specific event case studies where the subnet excelled or struggled. Such transparency will be important for attracting users (e.g. traders) to trust the subnet’s outputs.
Expansion of Markets and Queries (2025 and beyond): Going forward, Infinite Games aims to continuously expand the scope of what it predicts. The roadmap includes adding new categories of questions – potentially moving beyond the binary prediction markets into areas like forecasting crypto project metrics, macro-economic indices, climate or epidemiological predictions, etc. Because the system is general, it could be pointed toward any domain where an outcome will eventually be known. We might see collaborations with other platforms (for instance, integrating with decentralized finance protocols to predict metrics like TVL or yield, or with governance platforms to predict proposal outcomes). Moreover, the team has floated the idea of validators offering paid API access to the subnet’s predictions. This means third-party apps could query the network with custom questions (within certain limits) and get a probabilistic answer, paying a fee that is distributed to those who contributed to that prediction. Achieving this would effectively turn Infinite Games into a decentralized “oracle” service for foresight.
Long-Term Vision: The ultimate roadmap of Infinite Games is to become a self-sustaining, ever-improving prediction ecosystem. In practical terms, this involves growing the community of miners (to increase the diversity of models and viewpoints), scaling up the number of validators (to handle more queries and ensure robustness), and potentially integrating with the Bittensor governance mechanisms to hand over some control to TAO stakeholders. Since Bittensor as a whole is moving toward community governance, we may see Infinite Games subnet participating in that – e.g. via Senate proposals for parameter changes or funding grants for research. The project’s long-term success will be measured by whether its forecasts prove valuable in the real world. So a key ongoing goal is to demonstrate that the subnet can consistently generate predictions that beat baseline benchmarks (like simply trusting market odds or polling data). If it can, it opens the door to a new class of applications built atop Bittensor where AI agents continuously monitor and predict real-world events. In summary, the roadmap ahead is about scaling up (more data, more models, more users) and proving value (through benchmarks and real-world use cases), all while maintaining the decentralized, open ethos of the network.
The Infinite Games subnet had a clear launch and has articulated plans for future development. Here is an outline of its roadmap and anticipated milestones:
Mainnet Launch (Q2 2024): After an extensive period of testnet experimentation and refinement, Infinite Games officially launched on the Bittensor main network on June 20, 2024. This launch introduced the core prediction market mechanics (with Polymarket and Azuro event integration) and the initial Brier scoring mechanism for miner rewards. The subnet went live with a set of active validators and miners, effectively bootstrapping the world’s first decentralized LLM-driven forecasting system. (Prior to mainnet, the team ran closed tests and perhaps community test competitions to tune the parameters and ensure stability.)
Short-Term Improvements (Q3–Q4 2024): In the months following launch, the team focused on broadening the data sources and improving the quality of predictions. The roadmap included integrating a wider array of event streams: beyond Polymarket and Azuro, this could involve other on-chain prediction markets or oracle feeds (for example, real-world event data, additional betting platforms, or even news APIs). The goal is to maximize cross-domain expertise by feeding the predictor models more diverse information. During this phase, the team also planned to enhance the “guessers” (predictor models) themselves. One initiative mentioned is a community competition to attract better LLM-based predictors. By launching an open competition or bounty program, Infinite Games intends to encourage data scientists and AI enthusiasts to fine-tune models that can beat the current ones, thus continuously improving the subnet’s forecasting accuracy. Alongside this, there is work on developing a more user-friendly interface and comprehensive documentation – lowering the barrier to entry for new miners/validators and making the system easier to understand for outsiders.
Scoring Mechanism Upgrade (late 2024): A significant technical update in the roadmap was the introduction of the peer-based scoring system to replace (or augment) the initial Brier score approach. This was implemented to address some edge cases and to further incentivize useful contributions. By late 2024, Infinite Games transitioned to using peer comparison for scoring forecasts, which was an important milestone in optimizing the reward scheme. This change likely came after observing the subnet in action for a few months and identifying ways to better reward true insight (for example, rewarding contrarian but correct predictions more strongly). Such an update demonstrates the “self-improving” aspect of the subnet – not only are the models learning, but the subnet’s rules themselves are being refined over time.
Benchmarking and Results (early 2025): On the roadmap, the team has indicated an intention to publish performance benchmarks for Infinite Games, comparing its forecasting accuracy against industry standards and other forecasting methods. This “Big Benchmark” would be the first comprehensive evaluation of how well the decentralized network of LLM predictors performs relative to, say, traditional prediction markets, expert human forecasts, or academic models. Releasing these results would validate (or inform adjustments to) the approach. We can anticipate a report or paper from the team detailing metrics like Brier score averages, calibration of probabilities, and perhaps specific event case studies where the subnet excelled or struggled. Such transparency will be important for attracting users (e.g. traders) to trust the subnet’s outputs.
Expansion of Markets and Queries (2025 and beyond): Going forward, Infinite Games aims to continuously expand the scope of what it predicts. The roadmap includes adding new categories of questions – potentially moving beyond the binary prediction markets into areas like forecasting crypto project metrics, macro-economic indices, climate or epidemiological predictions, etc. Because the system is general, it could be pointed toward any domain where an outcome will eventually be known. We might see collaborations with other platforms (for instance, integrating with decentralized finance protocols to predict metrics like TVL or yield, or with governance platforms to predict proposal outcomes). Moreover, the team has floated the idea of validators offering paid API access to the subnet’s predictions. This means third-party apps could query the network with custom questions (within certain limits) and get a probabilistic answer, paying a fee that is distributed to those who contributed to that prediction. Achieving this would effectively turn Infinite Games into a decentralized “oracle” service for foresight.
Long-Term Vision: The ultimate roadmap of Infinite Games is to become a self-sustaining, ever-improving prediction ecosystem. In practical terms, this involves growing the community of miners (to increase the diversity of models and viewpoints), scaling up the number of validators (to handle more queries and ensure robustness), and potentially integrating with the Bittensor governance mechanisms to hand over some control to TAO stakeholders. Since Bittensor as a whole is moving toward community governance, we may see Infinite Games subnet participating in that – e.g. via Senate proposals for parameter changes or funding grants for research. The project’s long-term success will be measured by whether its forecasts prove valuable in the real world. So a key ongoing goal is to demonstrate that the subnet can consistently generate predictions that beat baseline benchmarks (like simply trusting market odds or polling data). If it can, it opens the door to a new class of applications built atop Bittensor where AI agents continuously monitor and predict real-world events. In summary, the roadmap ahead is about scaling up (more data, more models, more users) and proving value (through benchmarks and real-world use cases), all while maintaining the decentralized, open ethos of the network.
Keep ahead of the Bittensor exponential development curve…
Subnet Alpha is an informational platform for Bittensor Subnets.
This site is not affiliated with the Opentensor Foundation or TaoStats.
The content provided on this website is for informational purposes only. We make no guarantees regarding the accuracy or currency of the information at any given time.
Subnet Alpha is created and maintained by The Realistic Trader. If you have any suggestions or encounter any issues, please contact us at [email protected].
Copyright 2024