My AI Learning Journey

In this presentation I’ll talk about my deep dive, over the last two years, into AI and large language models (LLMs). My study has ranged pretty freely in response to my curiosity, more than any immediate attempt to make practical use of it. In addition to the usual prompt-engineering experiments, I’ve dabbled in fine-tuning small models. I’ve also studied the underlying technology from text books and research papers (brushing up on my linear algebra and Bayesian probability along the way!) My primary interest is incorporating the technology into software systems, but my intuition is that this is a field where deeper understanding will be very useful in non-trivial applications.

My policy when learning a new area, whether functional programming or deep learning, is to avoid carrying too much of my baggage at first. I don’t go in thinking “where are the bounded contexts and the value objects”! I want to learn it on its own terms. Later, the connections come unforced.

You won’t drown in math during this talk, but there will be a little, because I want to avoid wasting your time on metaphors. I won’t make a complete survey of what I’ve studied. More of a sampler of a few areas that excited me in my learning. This includes, for a start, multi-headed attention (“transformers”) and whatever is going on in “latent space”. This is not a detailed tutorial, of course, so you won’t come out with a full understanding of transformer architectures or the other topics, but I hope it will increase your excitement about why this technology is well suited to people like us who like tackling complex domain problems with abstractions and language.

And, yes, I will discuss a few connections with DDD that occur to me, still tentatively.