Hamel Husain breaks down why LLM products are so hard to build using the 3 Gulfs framework: gaps between data, models, and pipelines that each require different skills and mindsets to bridge. Developers have to wear different hats to connect these three islands and make the whole thing work.
People just cannot build AI products, even ourselves. We struggle to do it. So why is this so hard? And a conceptual model that I and my colleagues at Berkeley actually have come up with, very illustrative here, is this 3 Gulfs model of challenges that people have to face when they're building AI products. So there's a lot of text on the slide. I encourage you to visit it later. But the high-level idea here is that you are a developer. You are trying to build an LLM pipeline. And that LLM pipeline has to work on data. And in order to make your whole thing work, somehow you have to bridge these gulfs between these, to use a metaphor, islands. Developers have to wear different hats to bridge each one of these gulfs, and that thereby makes it very challenging. Each gulf, in other words, represents some mindset or some set of skills that you have to engage in. So let's talk first about
