Here are the detailed show notes for the podcast episode.
Episode Show Notes: Hash Rate - Ep 144 - SCORE Subnet 44
This episode reveals how SCORE Subnet 44 is transforming from a niche sports AI project into a scalable enterprise solution, leveraging decentralized vision intelligence to compete with human operators and generate real-world revenue.
Introducing SCORE: Decentralized Vision AI
- Max Sebi, founder of SCORE Subnet 44, introduces his company as a vision AI research firm focused on visual intelligence, contrasting with the text-based intelligence of models like ChatGPT.
- Computer Vision is explained as the technology that allows machines to interpret and understand the world through pixels from cameras. Max frames SCORE's mission as making every camera intelligent.
- He draws a parallel to Tesla's Full Self-Driving (FSD), which is a prominent example of computer vision, but clarifies that SCORE aims to apply this technology across many different domains.
Initial Focus: A Proving Ground in Professional Sports
- SCORE began by focusing on sports, specifically soccer, due to the team's background in computer vision, quantitative finance, and connections to the sports world.
- Max explains that sports provide an incredibly complex environment for vision AI, making it an ideal vertical to develop robust, state-of-the-art models.
- The subnet's goal is to provide elite-level AI analytics to sports organizations that are not at the top tier, such as their partner, the English football club Reading.
- Because of Bitensor's decentralized structure and competitive incentive mechanism—where miners compete for daily TAO rewards—SCORE can offer its technology at a fraction of the traditional cost. Max states, “Because of Bitensor's efficiency, let's say, we're able to provide them the same tech, state-of-the-art, and cut the price by almost 10-fold.”
How the AI Works: From Data Extraction to Strategic Reasoning
- The SCORE subnet operates in multiple layers to provide value to clients like sports teams.
- Data Extraction: The community of machine learning engineers mining the subnet provides AI models that extract raw data from video footage, such as player coordinates and movements on the field.
- Event Recognition: A higher level of analysis involves identifying specific in-game events, like a touchdown in the NFL or a goal in soccer.
- Reasoning and Insights: The most advanced layer, which Max identifies as their ultimate goal, involves reasoning. This allows the AI to describe events as they happen and extract strategic insights, effectively acting as an AI assistant coach that can suggest plays or identify player recruitment opportunities.
The Pivot: Scaling Beyond Sports with a New Enterprise Model
- The team realized that the complex models built for sports could be applied to numerous other industries. This insight prompted a 30-40 day pause to re-architect the subnet for scalability beyond its initial soccer-centric design.
- To prove their technology's value in new verticals, SCORE developed a unique go-to-market strategy called the "60-day trial pipeline."
- They approach Web2 companies with video-heavy, human-driven operations and offer a trial where the SCORE subnet competes directly against their existing human operators. If the subnet performs as well or better, the company converts to a paid customer.
Solving the Enterprise Data Privacy Problem
- A major challenge for decentralized AI is convincing enterprise clients to trust an anonymous network of miners with their private data.
- SCORE solved this by creating a dual-track subnet architecture:
- Public Track: A fully open-source track where miners process public sample data. Models are uploaded to Hugging Face, and inference runs on decentralized compute providers like Shush, ensuring complete transparency and verifiability.
- Private Track: An enterprise-focused track that uses TEEs (Trusted Execution Environments). TEEs are secure, encrypted enclaves that process data without exposing it to the underlying hardware operator, ensuring the client's private data remains completely confidential.
- This dual-track system provides the best of both worlds: the transparency of open-source development and the security required for enterprise adoption.
Major Announcement: Expanding into the Energy Sector
- Max reveals SCORE's first major non-sports enterprise client: one of Europe's largest petroleum companies.
- The use case involves deploying SCORE's vision AI across the company's network of automated gas station cameras.
- The AI will monitor activity to create an alert system, addressing issues like customers driving off without paying or causing damage. Max notes that the average resolution time for an issue at an automated station is currently 8-24 hours, a problem AI is perfectly suited to solve.
Tokenomics and Future Revenue Strategy
- SCORE is building an enterprise sales cycle where clients who complete the 60-day trial convert to monthly paying customers. The first revenue from a converted client is expected in December.
- A large portion of this revenue will be used for token buybacks to deliver value directly back to the subnet and its token holders.
- Max introduces the concept of "Alphanomics," inspired by Mog, and hints at a future model where token emissions are directly tied to incoming revenue. For every dollar of revenue, a dollar's worth of emissions could be "unlocked" for miners.
- The long-term vision is to use the data and models from these enterprise engagements to build a self-service, API-based platform, moving beyond the high-touch sales cycle.
Competitive Landscape and Ecosystem Perspective
- SCORE's primary competitors are not other vision AI startups but large, generic VLMs (Vision Language Models) like those from OpenAI or Nvidia. Max argues these models are slow, expensive, and lack the specialized accuracy of SCORE's expert models.
- The discussion shifts to the broader Bitensor ecosystem, particularly the impact of TAO Flow (TFuel)—a mechanism linking subnet emissions to token performance.
- Max explains that SCORE is not currently burning miner emissions because they are crucial for annotating data for their enterprise trials. He believes emissions should eventually be tied to revenue rather than being burned to manage market dynamics.
- He advocates for supporting both revenue-focused subnets and long-term research subnets like Nova, emphasizing that different models require different timelines and evaluation criteria.
Conclusion
SCORE's pivot to a revenue-first, enterprise model provides a tangible blueprint for other subnets to bridge the gap between decentralized technology and real-world business needs. Investors should monitor the success of their 60-day trial pipeline and revenue-tied tokenomics as key indicators of Bitensor's broader commercial viability.
