Over the next few posts, I want to work through stages to get us to a point where we have a growing understanding of how an LLM is trained, and how the inference (the normal use) process works. I am a firm believer that if you can’t explain something in a way that is easy to understand to most levels of technology-aware people, without dropping a load of jargon, you probably don’t understand the topic.

Now that I’ve said that, let’s get the jargon bit out of the way. By the end of this series of posts, you should be able to (if I’ve written this correctly) understand what a loss value is, how multi-dimensional mathematical spaces work, what a tensor is, how tokens work, what attention means in the context of LLMs, and what the inference process is, and what is definitely isn’t.

With the current hype around AI, there are two camps of pro AI/LLM people; those that deeply understand the technology behind it, and those that just want AI without knowing the underlying tech. I see myself firmly in camp A, and I believe that when others understand the technology, we can design creative solutions with LLMs as leverage. But, when we don’t understand the tech, LLMs become the proverbial hammer, and everything around the enterprise is a nail.

Just before we kick off with some learning, it’s important that you understand that I’m not anti-AI. You will see my passion for all things that call under the AI umbrella as these posts add up. However, sometimes when someone is in camp B (mentioned above), they can perceive someone like me explaining a solid (and the not-so-solid) use-cases of AI/LLMs as negative, or hating on the technology. It’s important to me that can see that my views are based in first principles thinking; and I want as many people as possible to understand the technology properly, so that we can build forward, in a cost effective and productive way.

That being said, let’s talk about apartments…