It's been some time since I have made notes on my "Gen AI" adventure. Here is an update in no particular order: THEORY The current wave of AI is based on "Large Language Models" (LLMs). Roughly speaking, these are models of the human language and how words (tokens) relate to each other. The LLM reflects the vocabulary of the particular AI application. Many databases of pictures are "tagged" or described in words the AI "understands" because the words are in the LLM. To this, add the magic of graphics cards, and you will have the ability of AI to find pictures that "look like" any given picture. Given these pictures, you have the associated words or tokens, which can then be matched to the "prompt." Ultimately, you have a two-way conversation about pictures mediated by the AI's vocabulary (the LLM). The key idea is that the words AI uses to describe a picture are automatically part of its LLM, so they can be used to cr...

This post gives me a glimpse into a typical morning with half a dozen AI tools at my elbow. The theme is "partnership": a flow of ideas and information back and forth between me, my human contacts and AI assistants. Grammarly made this paragraph a bit more readable. Grammarly is a writing assistant that is now so common we forget that its ability to "understand" and improve English would have been regarded as "AI" two years ago. Once such things become familiar, the goalposts are moved, and Grammarly is no longer considered "AI". In the following, Grammarly constantly corrects and re-words every sentence, hopefully for the better. I spend some time learning from a member of a Facebook group I moderate, tweaking his prompt: "a woman with a contemplative expression,. Her gaze is directed downwards, and she seems to be lost in thought. The background is a complex tapestry of intricate designs and patterns, blending both organic and geometric sh...