This is a great example of where the LLM (just like a junior programmer might) was starting down the path of implementing a change for a particular *page* and I had to redirect it to implement a more generic, long-term approach.

When thinking about how/if/when to use LLMs for context (especially at scale), consider these cost drivers. hobbs.link/46rGrYK

I like the title from The Economist: Can you make it to the end of this column?

Remember these are two things:

* Chat as a UI
* The underlying primary source of what is displayed

Yes, the ChatGPT app combines both.

But these do not need to be tightly bound.

Other options: Chat with data, or Graphical UI where the primary source content is an LLM.

Thinking about getting back into amateur radio. Transmitting in Hell Mode is not something I remember.

Slack's AI search is often quite useful. This example query and response was quite helpful (based on conversations spanning over a year with a large gap in between).

I asked Mistral's Le Chat if either Le Chat or Le Platforme (API) use inputs or outputs to train models. Le Chat responded unequivocally that it does not use inputs or outputs to train models. Their web page says same thing. But their current privacy policy says otherwise.

I asked Google "What is Content Chimera". Couple observations: it heavily prioritized something I recently posted on LinkedIn about a feature I was working on (but not yet released), and it made a semantic mistake (equating my company with the tool), but overall not bad summary.

This is an example of a useful LLM-driven auto-complete: what I was going to type anyway, but nice how it looked at the context to suggest the completion — and it saved some time.

In the end, I think my highest biz value is helping orgs:

1. Thinking bigger, while
2. Creatively defining structures to pull it off

The can only happen early, before potential implementors are involved. Don't automatically limit impact by sticking to existing silos.

Been reflecting on how spreadsheets probably are fine for content inventories in most cases. In some cases (especially complex sites) it makes no sense. But there's a tricky set of cases where teams decompose a problem just to make it doable with a spreadsheet.

The Build It and They Will Come fallacy
hobbs.link/4bVOZJK

Working on automating Content Assertions. An assertion is stating something about content that should be true, and these can be tested over time. Here I'm using it to work out the content transformation logic, to make sure old content doesn't break with new transformations.