On writing and LLMs
When I was younger I hated writing. I was bad at it in every way: I could not organize my thoughts. I lacked the vocabulary. I had nothing to say. And yet, as of late, I find an urge to write. Not for any particular reason, but it seems like I finally have something say, I have the vocabulary to say it, and I can (at least I think) organize the ideas a little bit. Perhaps this is simply a product of aging.
This writing bent feels very timely, too, given the rise of LLM AI workflows. I'm finding interacting with Claude on the command line to be fairly effortless, and I wonder how those without a solid command of the English language are getting by. It does help to have a large vocabulary. While Claude sometimes does play fast and loose with language (to the point of inventing its own terminology), it does seem to grasp some of the nuance when precise nouns and verbs are used.
Yet I can't help but shake the feeling that conversing using language is horribly inefficient. I should pause here and mention that I have always hated reading, and still do. Not just reading literature, but all reading, including technical matter. It is just so tedious, perhaps because I'm not a fast reader. The act of reading, from my software engineering perspective, is equivalent to deserializing a data stream. It is universal in some way, as serialization is capable of representing any information, but almost none of the information I want to consume is naturally expressed in serialized form. To me, imagery, process, structure, and hierarchy are the things I most want to understand, and these are things that do not map well onto a serialized sequence of tokens. And yet, this is how we interact with AI today: through the inherently serial medium of natural language.
I don't expect this... shall we say, convention, to last, but it is the dominant mode of interaction today. I remember before the advent of Google, back when we searched the Internet using Yahoo, Lycos, Hotbot, Altavista, and the like, that we were told the importance of understanding the logical operators AND, OR, and NOT when used in search queries; that somehow boolean logic was a basic skill required to effectively use the Internet. And where are we today? Is language fluency the new boolean logic? One can only hope.