The T-Rex Dilemma
Sharing AI Models Amidst Neck-Breaking Evolutionary Changes
The Problem of Redundant Downloads
We’re all exploring how to make client-side AI more efficient and effective, but there’s a serious pain point: redundant AI model downloads. Think about it—every time a user visits a different website, they might be downloading the exact same foundational model, wasting time, data, and storage. It’s a huge drag on performance and the user experience.
Our vision is to explore solutions where websites could use a wider range of shared AI models. Imagine a future where you don’t have to download the same model multiple times, no matter what site you visit. This isn’t just a technical plumbing problem; it’s a massive coordination challenge.
The Innovation vs. Stability Challenge
Building a shared ecosystem for AI models means balancing two competing ideas. On one hand, you have the “if it ain’t broke, don’t fix it” mindset. Developers might be hesitant to switch to a new model if their current one works well.
"It could mean sites agreeing on a small set of common models on an ongoing basis, or developers making compromises—perhaps using a widely available, slightly less preferred model rather than forcing users to download an entirely new one."
On the other hand, AI innovation is moving at a breakneck pace. The models we use today might be outdated in a year. How do we push the ecosystem forward with newer, more capable, and more efficient models while still maintaining stability? This also brings up complex questions about lifecycle management, like how to handle fine-tuning (e.g., refreshing LoRAs) when the underlying shared models are upgraded.
We Asked for Your Insights
Your experience is invaluable, whether you’re building with built-in browser AI, using libraries like TensorFlow.js or transformers.js, or deploying AI features across multiple sites. To better understand these challenges, we asked for your input through a series of surveys.
- Managed Devices Magic (For those managing AI deployments in an enterprise or educational setting)
- Multi-Site Mastery (For those overseeing AI features across a portfolio of websites)
- Integration Insights (For those developing widgets, libraries, or embeds with AI)
- Pains & Opportunities (For general insights on client-side AI challenges, including model downloads and storage inefficiencies)
Note: All surveys are now closed.
Thank you to everyone who contributed! We’ll be sharing our findings and next steps in a follow-up post soon. Let’s work together to build a more streamlined and efficient Web AI ecosystem. 🤝