Why the Future of AI May Be Smaller Than You Think | Jeffrey Li

What if the future of AI is not just bigger models in bigger data centers, but smaller, more specialized systems running directly on the devices you already use every day? In this episode of AI-Curious, we talk with Jeffrey Li, COO of Liquid AI, about the rise of small language models, on-device AI, edge AI, and specialized AI models. As the AI industry keeps pushing larger foundation models, more GPUs, more energy, and more cloud infrastructure, this conversation explores a different path: AI that is faster, more private, more efficient, and better matched to specific use cases. We dig into why not every AI task needs a giant cloud model, and why companies may increasingly turn to local AI, private AI, and domain-specific language models instead. For enterprise AI teams, the problem is not just technical capability. It is also AI ROI, latency, privacy, and deployment cost. If every query has to travel to a data center, the economics of AI can break down quickly, especially as AI subsidies fade and the real cost of large-model inference becomes harder to ignore. We also explore Liquid AI’s roots in MIT research, the origins of liquid neural networks, and how this work is shaping a broader vision for AI infrastructure. From Shopify search query rewriting to Mercedes in-car assistants, Jeffrey explains how small AI models and edge inference can make real-world AI products more practical and scalable. We also get into proactive AI, offline AI, AI agents on-device, and what it means to build AI that can run without depending on the cloud for every interaction. If you are interested in on-device AI, edge AI, small language models, specialized AI models, local AI, enterprise AI, private AI, AI efficiency, AI agents, and the future of AI infrastructure, this episode offers a practical and forward-looking look at where AI may be heading next. Topics include on-device AI, edge AI, small language models, local AI, AI agents, enterprise AI, liquid neural networks, private AI, offline AI, AI ROI, and specialized AI models. Guest Jeffrey Li — COO, Liquid AI Chapters 2:59 — Jeffrey’s background, from MIT to Snapchat, Scale AI, and Gather 6:56 — Why the “bigger is better” AI thesis is incomplete 7:26 — Type 1 vs. Type 2 thinking, and how that maps onto AI 10:27 — Why AI should run beyond the data center 13:02 — The ROI problem with sending every AI query to the cloud 15:58 — Why the economics of large-model AI may get harder 18:38 — Why many companies are moving toward smaller, specialized models 20:38 — The energy and infrastructure limits of large-scale AI 22:48 — Liquid AI’s MIT roots and the origins of liquid neural networks 27:26 — Real-world use cases for on-device and edge AI 29:38 — What proactive AI could look like in cars, phones, and homes 35:23 — How local models can give users more privacy and control 37:20 — Why smaller models can help in regulated, high-stakes use cases 39:56 — The Mercedes partnership and the future of in-car AI 43:12 — Why offline AI matters, and what it means to run directly on-device Follow AI-Curious on your favorite podcast platform: Apple Podcasts: https://podcasts.apple.com/us/podcast... Spotify: https://open.spotify.com/show/70a9Xbh... YouTube: / @jeffwilser All Other Platforms: https://www.buzzsprout.com/2230097/fo... For anyone interested in Jeff’s AI Workshops for their company: Reach out directly at [email protected]