As AI dominates tech headlines and corporate strategies in 2024, an important distinction is being blurred - the difference between developing AI versus consuming AI services. This mischaracterization risks confusing the market and overselling capabilities, but that isn’t anything new right? Take zero-trust, cloud computing, or even take a look back at the early-2000s with the web revolution. Urs Baumann tossed out a great questions in the Network Automation Forum - Slack recently, and I thought it would make for a good blog (or maybe venting session depending on how you look at it). Remember, these are opinions.
I think this is an excellent question. The moment Urs posted this in the chat, it made me think about medicine. There’s a clear difference between researchers that are advancing medical science and those practicing medicine. A family doctor doesn’t claim to be “doing pharmaceutical research” when prescribing medication. Let’s dig down a little further.
Core AI development represents a deeply interconnected ecosystem where each component builds upon and enables others. This is the story of our lives in tech, right? Training large language models requires novel neural architectures, which themselves emerge from fundamental machine learning research. This research, in turn, depends on insights gained from model training attempts and optimization techniques. The optimization methods then get refined. If you are doing AI, then you probably fall into one of the four cycles: