The Future of AI is Specialized, Why General Models Aren't Enough
summary
Argues that Web3 needs domain-tuned models not generic LLMs, showing a pipeline: tokenized model proposals, contributor-curated datasets, LoRA fine-tuning adapters, evaluation via chat tests, and deployment on an Open Lora framework that multiplexes many fine-tuned adapters on a single GPU. Claims per-inference proof of attribution on-chain lets platforms reward data contributors and trace which data caused a rogue or superior action. Gives concrete examples: DeFi sentiment models, Move and Solidity code-auditing models.
AI-written summary · about
chapters (14)
- 0:00Introduction & Talk Overview
- 0:13The Rise of AI & Everyday Applications
- 1:30Why Web3 Cares About Specialized AI
- 2:43General Models vs. Specialized Models
- 3:18Current State of Agents in Web3
- 4:24The Road to Proactive, Autonomous Agents
- 5:00Building Specialized Models on Open Ledger
- 6:19Data Contributors & Tokenized Model Proposals
- 7:29Fine-Tuning Workflow & Lora Adapters
- 8:37Model Evaluation & Chat Testing
- 9:20Deployment with Open Lora Framework
- 10:15Proof of Attribution & Rewards for Data
- 11:02Real-World Use Cases & Future Vision
- 11:41YouTube Analogy & Conclusion
related talks
32mLLM Future Is Multi-Model | Erik Davtyan #JSConfAM25
54mThe Future of AI: Building the Most Impactful Technology Together
32mSponsored Session: Beyond the Buzz: Navigating AIs Myths and Realities
10mWhy the Future of AI Depends on Blockchain
01h 01mSoftware Architecture for Gen AI
14mReliable Agentic AI, the Open Way
24mSoftware 3.0 and the AI Engineering Landscape
27mThe Cutting Edge of AI: From LLMs to Multimodal Intelligence