The Houston Rockets recently unveiled “ClutchBot,” an innovative generative AI mascot developed through a partnership with Theta Labs, signaling a significant step for crypto technology adoption in professional sports.
This collaboration represents the first instance of an NBA team leveraging artificial intelligence this way to enhance fan interaction. The initiative aims to redefine how fans connect with the team, offering real-time updates, statistical insights, and ticketing details via the Rockets’ official website and application.
The AI’s foundation lies in Theta Labs’ decentralized EdgeCloud infrastructure, utilizing advanced AI to deliver personalized, continuous support for fan questions. It’s specifically trained on Houston Rockets and NBA data, ensuring accurate responses from simple queries about game schedules to complex historical statistics and player information.
Mirroring the team’s dynamic play on the court, the Rockets are venturing into the future of fan engagement. ClutchBot operates on Theta’s Layer-1 blockchain and decentralized GPU network, providing rapid and secure interactions supported by over 80 PetaFLOPS of computing capability. Featuring sophisticated NLP models, API integrations, and dynamic fine-tuning, ClutchBot introduces a new era of interactive sports entertainment.
Below, Mitch Liu, CEO of Theta Labs, shares insights into the technology driving ClutchBot.
Q: How does ClutchBot’s natural language processing handle complex fan questions about things like historical data or player comparisons?
Mitch Liu: ClutchBot employs cutting-edge NLP methods, such as agentic AI workflows and large language model fine-tuning using real-time, domain-specific data. This enables it to understand user intent, fetch relevant information, and generate precise, coherent answers swiftly.
Q: What were the main obstacles in integrating API data with the Rockets’ internal databases for ClutchBot, and what other difficulties arose?
ML: Merging API data from the Rockets’ proprietary sources presented notable hurdles. We needed to construct a resilient data aggregation system to consolidate and align real-time data from various APIs, which often had different formats and inconsistent details. Minimizing AI hallucination was also crucial, especially concerning vital game, team, and player facts. We utilized sophisticated prompt engineering and a continuous model fine-tuning system to rapidly integrate new data as it becomes available.
Q: What specific function does Theta’s EdgeCloud fulfill in powering ClutchBot? Why opt for EdgeCloud over established centralized cloud providers?
ML: EdgeCloud acts as an intelligent orchestration layer, directing specific AI tasks to the most suitable and economical compute resource. Not every request demands high-end A100 or H100 GPUs; many can be handled efficiently by commercial-grade Nvidia 3090/4090 GPUs operated by our Theta community members. These operators earn TFUEL token rewards for their contributions, running on the underlying Theta blockchain. Our partners gain flexibility and crucial peak-time scalability, essential for live sports events, often at a better cost than centralized alternatives.
Q: How do you plan to gather user feedback to enhance ClutchBot and address issues after launch, particularly looking ahead to future NBA seasons?
ML: The AI refinement process is ongoing behind the scenes. In the future, we plan to incorporate direct user feedback, potentially incentivized through user engagement tokens. As our network of sports partners grows, we gain efficiencies in troubleshooting, as solutions developed for one partner can benefit all. Ultimately, this leads to a significantly more accurate and engaging conversational AI agent.