The Hex Update: Issue 029

the hex update
media
The Washington Post’s use of LLMs to improve article metadata; the value in broadcaster’s archives and metadata; some guidance on working with teams to set metrics; a story about the creator of the earliest surviving animated feature film; and some other cool things I came across this week
Author

Collin K. Berke, Ph.D.

Published

June 14, 2026

Let’s catch up

Welcome to Issue 029.

A pretty chill weekend. They never seem long enough, do they? Nonetheless, I’m excited for the start to a new week.

Let’s jump in and learn something new.

Three topics recently caught my attention:

For a bit of fun, I share an article about one of the oldest surviving animated feature films. It’s a really interesting story about a creator.

You’ll also find some additional links to other items I found cool this week.

Three things

Here’s what caught my attention this week:

How The Washington Post Scaled LLMs for Taxonomy Classification

This article overviews The Washington Post’s development of a scaled taxonomy classifier using LLMs. It provided a look into new developments around how the paper uses LLMs to apply a taxonomy with five schemas (i.e., Subject, Person, Company, Organization, and Geography) and around 20,400 taxonomy entries. To do this, it details how metadata tagging happens for Subjects (i.e., what the article is about) and Entities (i.e., who the article is about) using two different processes. Alongside how they did it, the author of the article provides some input on what doesn’t work, and they also detailed the real business impact such work has. This quote does a good job emphasizing the importance of this application:

metadata is what sits invisible beneath every article, shaping what readers see, what advertisers reach, and what editors understand about their coverage.

Why does this matter?

These types of metadata efforts have real business and operational impact. Metadata, although not seen by audiences, directly influences the reach and impact of content. This is also a great example of how LLMs can be applied to elevate the work of humans and improve media organization outcomes (i.e., revenue). Indeed, most media organizations will not have the resources to develop such a system internally. However, most media organizations can, at a minimum, have a strategy focused on improving the content metadata layer. LLMs can be helpful with this work.

How some broadcasters are turning archives into revenue with zero upfront investment using Veritone

This sponsored article features points made by Paul Cramer, managing director of media and broadcast at Veritone, during a panel discussion at a 2025 NewsTechForum event. Cramer’s points centered on the value inherent in broadcaster’s archives. This value cannot be realized without broadcasters taking steps towards enriching their metadata. Simply put, archives only result in value if they’re searchable and produce some form of revenue. Most media organizations, however, lack the funding and staff to complete such projects, as they can be costly in terms of time and money. The article highlights how Veritone’s products and services may be helpful in the completion of such projects. Alongside discussing the problem and solution, the article also emphasizes the use cases and benefits that could arise from metadata enrichment efforts. These include:

  • Improved strategic positioning to leverage AI
  • Clip licensing
  • Recommendation engine development
  • Licensing content for AI model generation
  • Ad intelligence reporting

Why does this matter?

Archives have value. This value is left untapped because of a lack of metadata that makes the content unsearchable. In my view, this goes beyond just licensing content for revenue generation. This type of work has a direct impact on audience’s ability to find and discover content. As mentioned with the previous Washington Post efforts, metadata enrichment efforts, or at the very least a strategy for improvement, are critical for media organizations.

The “art” part of setting metrics

Here is a post from Randy Au’s weekly newsletter, Counting Stuff. The post details challenges faced when analysts are tasked with helping teams set metrics for their projects. A key point was made: anything can be a metric, but what is rarer is the identification of useful metrics for teams. This involves significant work and deliberation, and it involves data folks being mindful of specific considerations when working through this process. The article described many of these, but one that caught my attention were considerations around the useful properties of metrics. To make this process more useful for everyone, the post suggested asking the following questions during metric development:

  • What does success actually look like?
  • What decision would this metric force you to make?
  • What’s the broader business goal we’re trying to achieve here?

The article also suggested keeping Goodhart’s Law top-of-mind.

In the end, the article pointedly mentions we have to pick something. I really liked this quote, which summarizes this point well:

For many projects, I don’t think there’s a singular correct answer to what the best metric is. The best I hope for is setting up a couple of sufficiently sensitive alarm bells that trigger when things need attention.

Why does this matter?

This article is a little different from what I normally include within these summaries. It makes some critical points, though. Media organizations rely on metrics to assess their efforts, better understand their audiences, and to better inform decision making across all operational areas. Each area has unique considerations when developing metrics, and Au’s viewpoints here provide some good guidance, especially when attempting to avoid vanity metrics.

Check out the links above for the full story in each item, not just my brief summary and analysis.

What I found interesting this week

I spent some more time in Google Dataform this week. This included reviewing some more docs and experimenting with the bq command line tool. I had to do a little detour and remind myself of all the different ways a table could be created within in BigQuery, though. This doc was quite helpful. As such, I sought to better understand how a table is created in a Google Dataform workflow using a .sqlx file. These docs were helpful to understanding how to perform this operation in Dataform. This includes the ability to create one of three table types:

  1. a regular table
  2. an incremental table
  3. a table view

The table type is defined in the config of the .sqlx file.

config {
  type: "table"
}

config {
  type: "incremental"
}

Cool things I recently bumped into

A collection of links to things I’ve found cool recently (or was reminded are cool).

That’s it for this update.

Go for it this week. It’s going to be a good one.

Cheers 🎉!

Let’s connect

If you found this content useful, please share. If you find these topics interesting and want to discuss further, let’s connect: