From: thepipeline_xyz
The Monad Madness event featured several teams presenting solutions leveraging Artificial Intelligence (AI) and Deep Learning. These projects span various sectors, from decentralized AI computation to healthcare data management and real-world asset tokenization, demonstrating the wide applicability and growing demand for AI-driven innovations.
Decentralized AI Protocols
Several teams are focused on addressing the scalability and accessibility challenges of AI through decentralized solutions.
42: Planetary AI Scalability
For over a decade, the team at 42 has been working on AI projects, including self-learning AI agents, conversational AI platforms, and training a fundamental model for spatial reasoning [00:24:52]. A persistent challenge they’ve encountered is the scalability of AI, noting limitations in API requests from centralized providers like OpenAI [00:25:08]. The industry currently estimates that trillions of dollars are needed to meet AI demand, requiring new nuclear power plants, data centers, and specialized chips [00:25:41].
42 proposes a sustainable approach by leveraging compute power already in the hands of users to create a “planetary layer” for AI scalability [00:25:56]. Their decentralized AI protocol is built on three pillars:
- Small Language Models (SLMs): Significant progress in SLMs, such as Qu 2.5, a 7 billion parameter model capable of running on a MacBook Air and surpassing GPT-4 in coding tasks by up to 4% [00:26:17].
- Excess User Compute: Modern MacBooks and PCs often utilize only about 20% of their power, especially those with Apple silicon, presenting an untapped resource [00:26:47].
- Swarm Inference: By combining SLMs with user compute, 42 enables a system where individual nodes work together to generate responses [00:27:10]. This involves a self-supervised inference where nodes assess their expertise to contribute, peer-review each other’s responses, and compile a final answer through a “Knowledge Tree” [00:27:24]. This approach aims to lower inference costs by 100x and dramatically improve AI accuracy by avoiding common errors through peer review [00:28:22].
The team behind 42 has a strong AI background, working with Transformer models since their early appearance and developing learning AI agents based on neural evolution of augmenting topologies [00:30:09]. Their CTO holds a PhD in Applied Mathematics with a thesis on distributed agents [00:30:42].
G: Economic Layer for AI and Compute
G is building the economic layer for AI and compute, aiming to create a new type of yield-bearing asset backed by real AI cash flow [00:33:42]. They identify compute as the “new currency” in the AI era, with GPUs being the asset capturing most value in the AI supply chain [00:34:04].
G tackles the illiquidity and high transaction costs of GPUs by financializing and tokenizing them, then providing liquidity through various DeFi use cases like GPU-backed stablecoins, lending, borrowing options, and futures [00:35:48]. Their architecture prioritizes transparency and decentralization, allowing for ongoing monitoring of GPU assets like uptime and performance [00:36:11].
The yield generated comes from actual demand on the AI cloud services market. One of G’s co-founders leads GMI cloud, a company projected to reach $100 million in revenue by year-end through selling AI cloud services to major companies [00:36:47].
AI in Data Management and Applications
Ever Network: Decentralized Healthcare Data
Ever Network is a decentralized healthcare data storage protocol built on Monad, addressing problems of siloed medical records and data breaches [00:17:09]. They provide users with ownership over their data, allowing them to decide where it’s stored, who accesses it, and who uses it [00:17:29].
A key feature is their chatbot, which allows users to “talk to that data” for symptom assessment without needing to consult a doctor [00:17:45]. Ever Network is also working with pharmaceuticals, insurance companies, and drug researchers to enable users to monetize their data, license it out, and get recruited for clinical trials [00:18:14].
JoJo World: Decentralized AI 3D Data Platform
JoJo World is a decentralized AI 3D data platform that collects high-quality spatial 3D data and provides it to companies like Google, Nvidia, and OpenAI for training text-to-3D models and large world models (LWMs) [01:01:46]. They emphasize that spatial intelligence is the future of robot training [01:02:01].
The platform incentivizes creators to contribute 3D data using a platform token [01:02:41]. Data buyers consume tokens to access this spatial 3D data [01:02:57]. The team notes that data collected from designers using professional software like Unity and Blender is of very high quality, crucial for training LLMs and LWMs, unlike lower quality data collected from cars [01:07:32].
Primus: Cryptographic Technology for Data Verification and Compute
Primus (formerly pX) aims to make real-world data usable on-chain by enabling verification and computation of any data in Web3 [02:21:43]. They address the limitations of on-chain data and the complexity of bringing Web2 data to Web3 securely [02:22:08].
Primus provides two core capabilities: data verification and data computation, utilizing advanced cryptographic technologies such as MPC (Multi-Party Computation) TLS, ZK (Zero-Knowledge) TLS, and ZK FHE (Fully Homomorphic Encryption) [02:23:22]. Their ZK FHE designs claim to be over 300 times faster than existing solutions like Zama [02:24:06].
Their short-term focus for Monad is on data verification capabilities, providing a standard protocol for developers to build digital goods marketplaces, verifiable data marketplaces, or prediction markets [02:24:47]. Long-term, they plan to deploy ZK FHE for confidential AI [02:25:24].
Score: Moneyball for Sports
Score Technologies is building “Moneyball for the World of Sports,” starting with football (soccer) [02:58:10]. They address the shocking reality that 98% of football clubs worldwide cannot access or utilize the valuable data they sit on [02:58:26].
Score’s solution involves AI models that watch games, extract all data, and fill in gaps with traditional stats [02:58:53]. They run the first decentralized network of football intelligence with over 200 AI models competing to provide insights [02:59:21]. Fans can contribute to creating new data using apps that teach machines about the game [02:59:36]. This enables possibilities like finding new talent, predicting tactical shifts, preventing injuries, and automating media content [02:59:48].
The company plans to open-source their models (like GitHub for football intelligence) and ingest 200,000 games per week [03:00:48].
Totem: AI DePIN for Daily Conversations
Totem is the first AI DePIN (Decentralized Physical Infrastructure Network) designed to unlock the value of daily conversations [03:39:53]. Unlike competitors focusing on low-value browser data, Totem targets the voice data market, estimated at over $100 billion [03:40:23].
Totem offers a device that attaches to a phone, capturing voice data and converting it into real-time transcription, translation, and summarization [03:41:13]. This data is then encrypted using ZK Proofs to become an anonymized dataset, which users can upload on-chain to Monad for tokens [03:41:26]. The device is designed by an Apple Hardware expert, featuring an acceleration chip and dual audio capture [03:41:47].
Their app provides structured notes and automatic summarization, ideal for professionals, students, and travelers [03:42:11]. The AI models, including Meta’s Llama, are optimized by experts from Google’s Gemini and Alibaba [03:42:35]. Users maintain control over their data with opt-in/opt-out options for recording and uploading [03:47:13].
Other AI/Deep Learning Mentions
- Paper Plane: Leverages AI agents for campaign management, recognizing that authorized on-chain data from user transactions can help build these agents [02:16:42].
- Talentum: Utilizes an AI agent to manage campaigns from start to finish, aiming to simplify the process for project owners [03:35:28].
Key Takeaways on AI and Deep Learning
Keone Han, CEO of Monad, offered insights relevant to AI and Deep Learning solutions:
- Skate Where the Puck is Going: Anticipate and predict secular trends driving technology use, such as the convergence of social media and betting, to incorporate into product design [03:52:52].
- Offer User Value: The core of any product should deliver ultimate value to the end-user. This intrinsic value, rather than just incentives or tokens, is what truly drives a flywheel effect [03:54:08].
- Short-Term and Long-Term Strategy: Builders should have a harmonious short-term strategy for initial momentum and operational excellence, alongside a long-term vision to unify the team, investors, and community [03:56:03].
These examples highlight the growing presence of AI and Deep Learning in the decentralized space, with a focus on addressing real-world problems through innovative technological and economic models.