Exploring the capability of LLMs

I am extremely fascinated by Large Language Models and what they are capable of. My first encounter with LLMs was about one half year before the launch of ChatGPT, when I was confronted with GPT-3 — and I was very unimpressed. ChatGPT changed that a little bit, but it took some months until I was invested more into my quest for exploration.

Today I experiment with the different models (LLMs and Image Generators), directly, for programming, and also as part of my own automations and programs — from a [private army of AI interns](/blog/ai-my-private-army-of-interns/) to structured analyses like a [Gemini-generated overview of tango-culture websites](/blog/geminis-overview-of-websites-about-tango-culture/), and most recently the serious tooling behind [TangoAtlas](/projects/tangoatlas-org/).

The arc of this project is basically the arc of the field: unimpressed by GPT-3 in early 2022, curious after ChatGPT, and since then treating the models as daily working instruments rather than toys. The practical results show up across the other projects here; this page tracks the exploration itself.

A good sample of the experiments lives in the genai-tagged posts on this blog.