Spectral’s Inference-Powered Web3 Vision

9 min readMay 20, 2024

Opening Remarks

Recent advancements in AI have been dominated by major tech corporations, leading to centralised and opaque systems. While open-source efforts like Mistral are emerging, essential ML tools remain largely under centralised control. The fusion of AI with Web3 technology offers a promising path to democratise AI, enhancing transparency and reliability.

Ethereum’s on-chain computing innovation faces network performance limitations, hindering full-scale on-chain AI processing. Spectral promises to address these challenges by introducing verifiable ML inference feeds and developing a Layer 2 solution optimised for on-chain ML operations.

Currently, most production-grade ML models are built by centralised entities using proprietary techniques, creating black boxes. This centralization clashes with Web3’s decentralised ethos, especially as AI Agents are poised to become mainstream. Closed ecosystems for on-chain AI Agents could obscure information origins and foster centralization.

Spectral bridges the gap between AI, ML, and blockchain, enabling the creation of on-chain Agents that operate on transparent, open-source knowledge bases through a common provenance layer known as the Inferchain.

Background

Spectral began as a Web3 startup focused on creating a credit risk assessment infrastructure that enables on-chain credit scores for DeFi decisions.

The company gained significant attention when Samsung Next joined a $23 million Series B round led by General Catalyst and Social Capital, with additional investments from Circle Ventures, Franklin Templeton, Gradient Ventures, Jump Capital, and Section 32.

On December 5th, Spectral announced the launch of its Machine Intelligence Network, aimed at generating high-quality, consumption-ready ML inferences and establishing the foundation for what they call the Inference Economy.

The team recognized the potential of merging Web3 technologies with AI and ML, deciding to pivot and use the funds raised to develop high-performance models. They offer substantial rewards and ensure tamper-proof, provably-fair model validation through a transparent, on-chain process.

This pivot is viewed by many as a smart and efficient move, given the market potential and gap. However, some see it as a quick decision to seize an opportunity, possibly starting a project without initially having the necessary expertise.

Hopefully, this article will help you decide which side you are on.

Project Overview

Spectral bridges the gap between AI, ML, and blockchain with innovative products enabling users to create on-chain agents using transparent, open-source knowledge bases via the Inferchain. Spectral offers two main products:

Spectral Syntax: A network where users create and monetize on-chain agents using a Solidity co-pilot that converts natural language into code. Users can also utilise community-created agents for Web3 tasks.
Spectral Nova: A machine intelligence network providing decentralised ML inferences to smart contracts. Nova incentivizes data scientists and ML engineers to develop and verify models for Web3 applications.

Agents from Spectral Syntax and inference feeds from Spectral Nova integrate through the Inferchain, enabling transparent, decentralised, and verifiable AI applications in Web3.

Key Actors in the Spectral Network

Creators: Web3 companies that post data science/ML challenges, set performance benchmarks, and offer rewards. They earn revenue from the use of inferences generated by their challenges.
Solvers: Individuals or teams that tackle these challenges. They can win bounties and earn the majority of the revenues from ongoing use of their model inferences by consumers.
Validators: Ensure model integrity and quality. They create unique test sets, verify solver responses, and check that models meet the performance benchmarks.
Consumers: Discover and pay for inference feeds that meet their data science/ML needs, integrating them into their applications.

Technical Design Principles

Privacy-Preserving Machine Learning: Spectral uses zkML to preserve privacy during training, evaluation, and consumption of ML models, protecting Solvers’ intellectual property and ensuring a secure, trustless network.
Technical Abstraction: Spectral simplifies complex cryptographic and mathematical concepts in zkML, providing users with a frictionless experience while maintaining the integrity of machine intelligence during inference.
Validator-Based Quality Control: Validators ensure the quality of machine intelligence by using a randomness beacon to verify models and evaluating performance with diverse, industry-accepted metrics. They openly disclose all intermediary steps for transparency and accountability.
Customizable Solutions: Spectral supports scalable and custom ML models, catering to both general use and specialized requirements.
Network Effects: The Machine Intelligence Network operates as a self-scaling flywheel, incentivizing Solvers to produce better models and web3 companies to post relevant challenges.

Incentives and Rules of the Network

Creators: Post new challenges to unlock additional revenue from multiple consumers who find their challenges relevant.
Solvers: Build better models across various challenges to earn rewards and preserve intellectual property, allowing for perpetual incentivization and fee sharing.
Consumers: Request inferences to harness the decentralised community’s power, obtaining high-quality, ready-to-use inferences at a lower cost than in-house development.
Validators: Validate inferences to earn steady revenue and secure earnings by staking.

Team, Foundational Support and Strategic Partnerships

Leadership and Funding

CEO & Co-Founder: Sishir Varghese, since June 2020, with a background in architecture and blockchain strategy, and experience at Gitcoin and Loopring. He is a Columbia University alumnus.
Funding: Raised $23 million in 2022 led by General Catalyst and Social Capital, bringing total funding to $30 million. Other investors include Samsung and Gradient Ventures, supporting Spectral’s mission in the Web3 credit score space.

Key Partnership

EZKL Collaboration: Integrating zkML into Spectral’s decentralised ML oracle network to enhance platform security and efficiency, enabling the creation and validation of zero-knowledge proofs for ML model predictions.

Vision and Launch

Spectral aims to provide a transparent, private, and secure source of ML inferences through a two-sided marketplace offering high-quality inference feeds for smart contracts using zkML to preserve privacy and protect IP.

It launched in November 2023 with a Web3 Credit Scoring challenge, attracting over 200 modellers, including 40 in the Early Modeler Program. Inference feed consumption starts in February 2024.

2024 Roadmap

Q1 2024

Launch Machine Intelligence Network (Alpha) and establish a base of early adopters.
Conduct ML challenges, deliver inferences, provide zkML verifications, distribute rewards, and start consumption of AI/ML feeds.
Launch SPEC governance and Challenge 2 (Solidity Code Generator).

Q2 2024

Scale network usage and build advanced features for Modelers and Consumers.
Launch permissionless Creator and Validator onboarding, Modeler Reputation Rankings, and secure the network using SPEC.
Web3 AI Agent for Solidity.

Q3 2024

Curate versatile inference feeds for diverse use cases.
Launch customizable inference feeds and solicit new ML challenges from multiple firms.

Q4 2024

Improve the speed, cost, and efficiency of on-chain feed consumption.
Launch InferChain on testnet in 2024, with a Mainnet launch in early 2025.

Tokenomics

Market Data (as of 15/05/2024)

Market Cap: $84.71 M (Rank #460)
Fully Diluted Valuation (FDV): $818.70 M (Rank #168)
Circulating Supply: 10.52 M (10.52% of Total Supply)
Total Supply: 100.00 M
Max Supply: 100.00 M

The SPEC token is essential to the Spectral Network, facilitating governance, staking, and value exchange.

Governance: SPEC holders vote on platform upgrades, propose changes, and influence governance rules, ensuring a democratic and flexible system.
Staking:

Spectral Syntax: Staking SPEC allows users to create and monetize Agents and pay for services with benefits like faster transactions and lower fees.

Spectral Nova: Validators and Solvers stake SPEC to ensure challenge integrity and high-quality model production.

Incentives: Staking SPEC provides rewards and fee discounts, aligning the interests of Solvers, Validators, and Consumers with the network’s success.

Competitors

Within the incypro AI sector, Bittensor stands out as one of the leaders (at least for now), with a market cap of approximately $2.3 billion. Bittensor and Spectral differ in several key areas. Bittensor focuses on creating a decentralised machine learning network that rewards contributions to a global model with its TAO token. Spectral, on the other hand, integrates AI, ML, and blockchain to form a comprehensive ecosystem for data processing, using the SPEC token for governance, staking, and transactions.

Technologically, Bittensor emphasises collaborative AI training through a decentralised neural network, while Spectral uses advanced technologies like Zero-Knowledge Proofs (ZKP) to enhance AI and blockchain integration.

This means Bittensor focuses on collaborative AI development and data sharing, while Spectral addresses AI integration, secure data processing, governance, and financial transactions.

Additionally, ML compute marketplaces, led by Bittensor, concentrate on ML model training and deployment, but new competitors are rapidly changing the market landscape. Spectral sets itself apart by enabling model secrecy and accuracy verification via zkML (at least in theory), adding another layer of innovation to its ecosystem.

Bullish Fundamental Factors

Spectral pioneers the fusion of blockchain and ML, with a focus on zero-knowledge machine learning (zkML) for a competitive edge.
Although Bittensor leads presently, there are gaps in the market ripe for seizing, especially considering that most zkML projects are yet to launch with a token.
The machine learning industry is currently dominated by centralised giants. The adoption of crypto might shift the interest towards decentralisation, which provides Spectral with a rich market opportunity.
If the integration of zkML really works, it revolutionises this space by allowing ML inference verification on-chain, keeping data and model specifics confidential.

Bearish Fundamental Factors

Adopting zero-knowledge proofs in ML via EZKL introduces significant computational demands and software development complexities. As of today, it hasn’t been integrated in an efficient and sustainable way.
The overhead and practicality concerns exemplify the significant performance implications of employing zkML.
Traditional ML can support large models within 32GB memory. In contrast, zkML for the same scale could necessitate terabytes to petabytes, far exceeding practicality for large models.
The AI space is highly competitive, with new projects launching frequently.
Spectral still lacks strategic partnerships and has to navigate through all zkML challenges mentioned.

Closing Thoughts

Spectral is pioneering the fusion of blockchain and machine learning, focusing on zkML for a competitive edge. While Bittensor leads, Spectral can exploit market gaps, especially as many zkML projects haven’t launched tokens yet.

Spectral’s Nova product, the first web3 oracle delivering AI inferences to DApps and smart contracts, could transform them into intelligent, self-learning agents. Spectral Syntax, an AI tool for generating Solidity code, further solidifies Spectral’s position in the AI-blockchain sector.

However, integrating zkPs in ML with EZKL presents significant computational and developmental challenges and the AI industry is competitive.

If successful, Spectral could revolutionise on-chain ML inference verification while maintaining confidentiality.

Disclaimer

This presentation has been prepared by Greythorn Asset Management Pty Ltd (ABN 96 621 995 659) (Greythorn). The information in this presentation should be regarded as general information only rather than investment advice and financial advice. It is not an advertisement nor is it a solicitation or an offer to buy or sell any financial instruments or to participate in any particular trading strategy. In preparing this document Greythorn did not take into account the investment objectives, financial circumstance or particular needs of any recipient who receives or reads it. Before making any investment decisions, recipients of this presentation should consider their own personal circumstances and seek professional advice from their accountant, lawyer or other professional adviser. This presentation contains statements, opinions, projections, forecasts and other material (forward looking statements), based on various assumptions. Greythorn is not obliged to update the information. Those assumptions may or may not prove to be correct. None of Greythorn, its officers, employees, agents, advisers or any other person named in this presentation makes any representation as to the accuracy or likelihood of fulfilment of any forward looking statements or any of the assumptions upon which they are based. Greythorn and its officers, employees, agents and advisers give no warranty, representation or guarantee as to the accuracy, completeness or reliability of the information contained in this presentation. None of Greythorn and its officers, employees, agents and advisers accept, to the extent permitted by law, responsibility for any loss, claim, damages, costs or expenses arising out of, or in connection with, the information contained in this presentation. This presentation is the property of Greythorn. By receiving this presentation, the recipient agrees to keep its content confidential and agrees not to copy, supply, disseminate or disclose any information in relation to its content without written consent.