Proof of Coverage Making the Internet AI-Native with Grass | Andrej Radonjic
Oct 6, 2025
Andrej Radonjic, co-founder of Grass and expert in AI infrastructure, discusses the transformative shifts in AI and crypto since 2024. He highlights the critical need for live context retrieval and how Grass is developing robust, scalable data solutions. Themes include diminishing returns in AI scaling, the potential of multimodal systems, and Grass’s innovative products like ClipTagger for video search. Andrej also outlines Grass's unique position to challenge existing internet giants with its vertical integration and real-time data capabilities.
AI Snips
Chapters
Transcript
Episode notes
Scaling Hits Diminishing Returns
- AI scaling in a single dimension faces diminishing returns and hits S-curve limits.
- Labs shift dimensions (e.g., MoE, test-time scaling) to regain meaningful improvements.
Multimodal Scaling Remains Open
- Multimodal models still benefit strongly from scaling because data is scarce and complex.
- Video and image models require vastly more data and different capabilities than text models.
AI Search Is The New Essential
- LLMs increasingly need web search to fill knowledge gaps and provide up-to-date context.
- Demand for AI search will only grow as models become primary interfaces to the internet.