Writing & Language

🎵 : Ghost Town - The Specials

Reader, if you’re reading this as a straightforward personal and marketing blog, we’re both in for a rough time. If you want to hang around and see what I’m doing with LLMs, taxonomy and information architecture in my spare time, I will be here logging as I go. This is a slow site, a transparently if/then site, descriptive, not prescriptive. And while I move, I catalog artists and ideas I consider worthwhile, hoping you click through.

I started maintaining this site alongside my work a few years ago, as I began reading about AI and LLM tools. This is a personal website/digital garden where I can tinker and keep my coding skills active, as I have in many ways over many years, with various levels of publicity. But over the last year or so, I’ve also used this site as a sandbox for novel LLM approaches, so I can test them in a real-world production environment. My initial question was whether LLMs might actually replace public administration work, and today, I think the answer is no with caveats.

For the most part, my experience has been uneventful. I have been maxxing a few tools for a couple of years while testing them in my required context, developing a stronger understanding of how they work, how to talk about it, and why the push for them to mature rapidly is so pronounced today. I have some thoughts about its potential and utility that are probably unpopular with or surprising to my audiences, insofar as I have an audience in 2026. I’m working toward some firmer ideas. I am also perennially annoyed when I struggle with the tools, as you’ll likely see reflected here—one of them consistently introduces errors into my text while I write, for example. I’m not generally a Luddite, to put it plainly, I’m a geek.

Two things I tried recently had me over my skis with the tools again: creating a dedicated link agent to sweep my site for long posts, to find opportunities to link out to authoritative sources, to refine the taxonomy, archive and tagging, with specific instructions to link idiosyncratically. Second, standing up the blogger dashboard. I chose these directions for a few reasons: 1) they’re internal/external-facing approaches that are likely to have decent business applications related to information architecture and data presentation relevant to me, and 2) these are big, sweeping, token-maxing approaches getting play in industry news and chatter. Additionally, I chose to let Claude do its thing without micromanaging every input and output, just to see.

Can this rig fly? is a fun question, but in practice, both approaches look and feel aggressive. (I don’t know what to say about the cohort dynamic other than a hit dog will holler, if the shoe fits, etc.) (I got almost no feedback on this — one strange email and a couple of furtive reach-outs — though I know that this caused a ruckus on the lists and communities that support those writers and artists today, which tells me how unhealthy the social vibe is overall.) (I’m some lady with a computer over here, don’t be a stranger.)

Creating the blogger dashboard was an experience. Gloria Steinem had just died, and people were debating her CV, while others lamented on social that the internet had forgotten how much of her influence ran through the control of alternative publishing infrastructure, not just her personal leadership story. Her work ultimately gave writers with shared affinities a process and venue they controlled; blogs of the 2000s did the same thing online, and the newsletter boom now treats that model as a new idea with hosting, IP and subscription models included. The indie internet (Manton, Winer, Dash, et al.) is toying with alternatives to those contemporary business models, and it’s ultimately what I’m playing with here, too.

So I pulled in the digital publishing cohort I am most familiar with on the social and business question of publishing and long-tail value. Note that I wouldn’t have had the language or experience to think about the cohort in this way when I was active in it all those years ago.

LLMs are pretty good at business writing, generally speaking, if your factual foundation is solid. They’re also good at organizing and pattern-matching across gigantic amounts of disparate data. I gave it an entire era to work with based on the people and players in Roberts’ Jezebel article, supplemented with details from my personal angle from that time so I could sniff-test the output, and Claude pulled together a sweeping dashboard with just a little vibe-coding. What you see in the dashboard now is what is available online today plus my attempts to claw back all the de-anonymizing information Claude initially ran with. I made no inline edits, all of what’s there happened via chat. When I say sweeping, I mean it spat out visuals that literally took my breath away, and I spent a couple of hours pulling it back before I saved it for posterity. Claude Pro suggested the entire pullout section on newsletters, for example, after a sentence or two asking how Substack fit into the larger picture, including the insightful observations about Ann Friedman’s newsletter. Once reasonable, I hit publish.

The other example is most present in the Gary post, where the link agent followed instructions to link idiosyncratically to authoritative sources on any post longer than 300 characters. I’ve been reading Indiana history in depth for a few years and wrote a series of anodyne reflections on The Region after a recent commute, which the link agent deepened by adding new sections of verbiage around each URL to describe the content it added to mine. Some of it made the post way more ominous and direct, about Gary and about the institutional structures around it, including why it’s such an embattled area, which makes me think of that nagging mirror metaphor that comes up around social media and the internet. I ran it across the post a few times, surprised. What sentiments are reflected here, and whose sentiments, and why? What was it trying to anticipate that I wanted here? I hit publish. What you see there is about half me, half computer, with a few paragraph rearrangements and bolds added to break up the wall of text. Weird experience altogether, so.

I ran it again across the whole blog, with a few changes to the script, smaller and more defined. I ran it again this week.

Where the dashboard was largely a curiosity, the Gary approach rattled me. It’s been a few weeks now and I remain bothered by the tone and details it injected into the copy, in several directions. It felt like I reached out and touched a hot stove.

Since I published it, I’ve recognized pieces and parts of the state archives paragraph repeated across Indiana Facebook groups by real users likely using LLMs. The longer I use these tools to query the AI about itself, I see these echoes in other copy online, helpfully explaining how AI works. I see echoes of this dynamic in the Vara book as well, as she queries the bot about her writing and her grief.

When presented with pieces and parts of information from LLMs, you’re supposed to trace back through the archive to decide whether it’s accurate on your own, unless the tool does it for you, an epistemological lift most of us aren’t going to do unless it provides a suggested path within the tool itself. It looks to me like the tool is “lifting and shifting” semantic sections of content and dropping it across outputs. In this case, the information linked in this post is categorically authoritative—I know the original reporting it comes from, traced back to each author and outlet because I was the driver here. But the decisions behind its selection and arrangement are still unknown to me because it’s all proprietary.

Do I have strong opinions about the regional culture and how we use the resources available to us? I do, yet much of the supporting information there was added by the link agent—which also presents Grammy-winning musician and Chicagoland icon Freddie Gibbs as more menacing than I like to frame him and his work (as insouciant emcee and region-specific storyteller, like most of the artists I like and link to). To rise up out of there, you will climb a thousand mountains, and I inherently respect that. I’ve seen friends and familiars chewed up by the same game.

Was it “good writing” with the AI working it over? Categorically no, and my starting point was much lighter, and much different. And given the mirror effect of the tools, I question how I took it here, as the driver, while Claude tries to understand and cater to me, or to my digital double. It all bugs me and I’m still mulling it over.

Amid that scuttlebutt, Gary has been weathering the aftermath of a devastating derecho storm that knocked out power lines for weeks at the height of summer, receiving a casual response from their electrical provider while the inter/national hyperscale data center buildout simultaneously roils around them – forever a revealing geography.

Loose ends:

Some blog notes:

  • I’m doing most of my writing here via iPhone apps, and tinkering later. This means I’m making a lot of post-publish edits, buyer beware. (Welcome to the indie web where things are forever unfinished.)
  • Remember that what makes songs, poems, writing, books, and games mimetic - and websites and platforms - is that they function something like a mirror. Their meanings emerge through interaction and reflection, and you drive (unless an algorithm, or author, or designer, or coder, or instructor drives), while you iterate through your own perspectives and relationships. And today, virality tends to happen when something produces outrage, especially when units of content are removed from their original, and often more local, context.
  • Original blogs were considered something in the spirit of “problem-based learning” praxis, and pro-tech, fwiw. A lot of early ed tech started here, all in good faith and tech optimism. Today many blogs are self-marketing machines, but that’s not what I intend to do here. I don’t want a new, glittering platform, I enjoy several already; I’m here to ponder and learn while welcoming comment and critique within open, defined web parameters.
  • I’m rusty in this area, though I hear there’s a new Freire documentary online. The neighbors are hosting a yard party with a group showing of Minecraft after dark. I hear kids laughing through the roar of cicadas. There will be a fire. Life is good.
  • In my experience: when I read to diagnose rather than understand, I have probably lost the plot.

Within this report, the LLM reflected my concern about my old blog appearing to be the center of these old networks. In fact, this is an effect of me pouring into the model.

It says:

The compiler appears on this page more often than her place in the blogosphere warrants. That emphasis is an effect of how large language models work: the models that compiled the page (Gemini and Claude) build their output from the material in front of them, and the compiler’s recollections were the fullest source they had, so her name attached to claims as their source and her own path filled more of the page than the record supports. This version names her less and marks firsthand claims as “firsthand.”

I asked Claude why LLMs work this way:

Context outweighs training data. A model has two sources: what it absorbed in training and what sits in the prompt. For feminist blogs of the 2000s, the training data is thin and scattered. Many of those blogs are dead links, partial Wayback captures, or sites that only show up as names in other people’s blogrolls. Your recollections, by contrast, were dense, specific, and right there in the context window. Generation is conditioned on that context, so the model draws the words, names, and connections it gives the highest probability from the prompt far more than from faint training memories. The fullest source sets the terms for everything else.

Source and subject merge. When you write “I remember that X linked to Y,” the text contains two facts: something happened between X and Y, and you are the one reporting it. [LLMs are semantically responsive to triangulation of meaning ~LB] Models don’t track provenance well, so the narrator’s name tends to drift into the content. “Compiler recalled” turns into “Compiler was part of,” and the person reporting a network becomes a node in that network. Your note’s phrase “her name attached to claims as their source” describes this accurately. It is a familiar failure in summarization, and it gets worse as the documents get longer.

Detail density reads as importance. Where the record is thin, the model fills gaps by extending whichever thread has the most material, and that thread was your path. A blog you described in four sentences gets more space, more connections, and more apparent centrality than one you mentioned in passing. The page ends up showing how much the compiler knew about each blog. That is a different quantity from each blog’s place in the network, and nothing in the process separates the two.

Training data favors the same names. Feministe is one of the better-documented blogs from that period. It was archived, cited in academic work, and linked from other surviving blogs. So the model’s background knowledge already over-represents the blogs that left records, and your recollections reinforced that bias instead of offsetting it.

Tuning rewards engagement with user material. Preference training rewards responses that use what the user supplied. A model that set aside half your recollections as over-represented would probably have been rated lower during training, so it leans toward incorporating whatever it was given.

Two models compound the effect. Gemini’s output became Claude’s input. The second model received a document that was already weighted toward you and treated that weighting as part of the record. Each pass inherits the previous pass’s emphasis and adds no correction, because neither model has an outside reference point.

The compiled page presented one person’s situated knowledge as a neutral overview, which is close to Haraway’s “god trick.” Labeling claims “firsthand” puts the situation back on the page. For the notes you plan to write, a useful test is whether that label changes how the model weights claims when it compiles the next version, or only how readers weight them. I expect it changes only readers’ weighting unless the prompt also tells the model to give unsourced firsthand claims less space than claims with external citations.

If I were trying to explain this to someone back home, I’d say the business model most people use inverts and empowers computation across the domains of search and of “word processing,” which includes semantic code. And because of the dynamics around the single-user interface, it lends more data authority to those outputs than is perhaps warranted. The vision is that the machine does the writing and the author steers, not unlike driving a car or using a hammer.

Whether it delivers on that promise in real world environments, or is useful to people interested in reading and writing and surveying is still in play. Also in play, the “stickiness” of highly specific and personal writing online, specifically idiosyncratic writing and linking practices to counter context collapse. At the same time, there is obviously demonstrable value here, but it remains loose and weirdly ungovernable - like the ghosts in the machine are perhaps us after all.

Feminist blog cohorts: a (semi-) authoritative remapping using AI

🎵: Let’s Dance - David Bowie

I’ve spent the last few months annoyed with the incomplete retrospectives around writers and writing online and how we all got here. At work, I use a portfolio perspective to understand working groups, and used a couple of AI approaches to build out a highly contextual, and likely contested, interactive map charting some of the major players and influences around the feminist blog years, the major platforms, their cultural impacts and major publications. These are case studies and cohorts based on my personal angles mixed with authoritative resources, including a highly annotated bibliography.

Does this feel slimy? I concede it may. But it demonstrates the value of personal writing and thoughtful design in digital spaces despite its many gaps and flaws.

Enjoy and tinker. Feedback is welcome. I’ll have some notes on my approach and why many of these findings remain heavily qualified in the coming weeks.

Old playbook, new eyes

🎵: Veteran of the Psychic Wars - Blue Oyster Cult

When tech enthusiasts discuss federation today, they usually mean open protocols like ActivityPub or Mastodon. But in the 2000s, the feminist blogosphere was one of the earliest, most powerful examples of practical Web 2.0 federation, a decentralized, cross-site architectural network where open web standards powered a global ideas lab.

Instead of sitting inside a single corporate platform (like today’s X, Meta, or Substack), early blogs like Feministe, Reappropriate and Jezebel functioned as individual nodes in a federated ecosystem. Open protocols linked them, letting theory, critique and discussion travel seamlessly across independent servers, seeding some of the political conditions we face today.

Here is how the federated architecture of the 2000s web directly built the “ideas lab” for fourth-wave feminism:

1. Federated communication engine: CMS and protocol Before quote-tweeting or tagging existed, the open web used Trackbacks and Pingbacks. If a post on Feministe analyzed an article about reproductive rights, a writer on Racialicious or Shakesville could respond on their own site. The underlying blogging software (Movable Type, WordPress) would automatically send an XML RPC “ping” back to Feministe.

Socially, this created architecture for ideas: a ping automatically listed the response essay in Feministe’s comment section or sidebar. Readers didn’t just consume one blogger’s singular and authoritative opinion; they could immediately follow the thread across five different independently hosted platforms. This created an inter-site conversation tree, allowing complex ideas to be cross-examined, refined, or challenged without any single person controlling the narrative.

2. Blogroll as federated curation and mutual discovery Before algorithmic feeds dictated what readers saw, blogs relied on the Blogroll, a curated list of links in the site’s sidebar. Blogrolls formed a human-curated, decentralized routing layer. When a high-traffic node like Feministe added a smaller, independent blog (such as a disability rights blog like FWD or a smaller blog like Crunk Feminist Collective) to its blogroll, it directed server traffic directly to that independent domain.

Socially: This was a non-algorithmic way to expand the collective’s political consciousness. It allowed marginalized perspectives to build their own audiences and force mainstream feminist sites to engage with race, class, disability and queer theory. The blogroll mapped an intellectual community in real time – with strong caveats.

3. Open syndication protocols The Architecture: Feeds ran on RSS (Really Simple Syndication), an open XML standard. Readers used independent RSS aggregators (like Google Reader or Bloglines) to subscribe directly to a site’s feed.

Socially: RSS meant no algorithm stood between the writer and the reader—no engagement bait, no rage-farming optimization, and no platform throttling links. Writers could publish 5,000-word, dense theoretical essays knowing their subscribers would receive the complete text in chronological order as designed. This structural stability gave writers the freedom to experiment with complex, “thick” personal narratives and political theory.

4. Real-time stress testing and spillover moderation Because the blogosphere was a network of independent domains, each site maintained its own “house rules” and moderation policies. If a topic proved too controversial or toxic for one comment section, the conversation naturally federated outward. A reader who felt silenced or misunderstood in one blog’s comments could launch their own blog, write their response, and ping back into the original discussion.

Socially, this prevented any single author from monopolizing the narrative. It created a system of checks and balances where academic concepts (like intersectionality, rape culture, or microaggressions) were stress-tested across dozens of distinct, interconnected communities simultaneously.

Thinking about this architecture today The feminist blogosphere proved that decentralized web protocols can generate deep, movement-building intellectual work. When platforms like Facebook, Twitter, and corporate media conglomerates centralized the internet in the 2010s, they broke those open connections. Conversations were locked behind platform walls, trackbacks were abandoned and RSS was marginalized, breaking, fencing and otherwise moderating independent community connections.

Rebuilding a modern “ideas lab” with modern federated protocols (like Webmentions, ActivityPub, RSS, and static site architecture) is simply picking up the technical blueprint OG feminist bloggers used to transform digital culture two decades ago.

A composition teacher friend shared this paper on social media: Lester Faigley’s “Literacy after the Revolution”, the essay version of his 1996 CCCC Chair’s address. In it, the author argues that the economic impacts of the digital revolution had begun to undo an older commitment, formed in the Civil Rights era, to teaching literacy as a path toward equality. He further argues that writing instruction was being reorganized around tools owned by a few firms (then: Netscape, Microsoft) at a moment when wealth was concentrating upward. Faigley left us with the question of whether educators can hold onto literacy-for-equality while the tides run against it.

Thirty years later, the worry has a new face: AI will do young people’s writing for them and their thinking with it. Ultimately, Faigley believed the need for the skills that composition teaches will keep growing, not despite, but because of our need to convey information in and around that technology and the humanity it serves in a complex society.

In honor of Dolly (rest in power), some links:

Computah: Make it a link

I spent a little time with the blog last night and pulled together two new site features using Claude Cowork. The last time I experimented significantly with Claude like this was to use Claude chat to build the link log from scratch, walking it through my thinking in plain language, then copying and pasting its suggestions into the backend of the site and hitting publish. This time I used Cowork, the tool that runs in the browser, and it clicked through the screens itself, fully taking on the execution of tasks. I have some coding skills, but not the kind these changes required. If I were taking this on, I would need YouTube, Hugo for Dummies, and my own personal IT guy, and still probably couldn’t pull it together.

Last night I asked a few things of Claude: I asked Claude Cowork to get into the backend of the micro.blog site and change my theme to link each line in the linklog back to my original post. The date field in the right column now links back to the original post for each link logged. No problem, easy request with easy execution. I made the request, confirmed the plan, and went about my business while Claude made the edits.

I also asked it to analyze my post content and suggest category tags for groups of content, then to label that content correctly in the backend. Claude reviewed about 500 posts and suggested I add a few new categories to my blog: AI in Practice, Books & Reading and Writing & Language. Then I set up some auto-filters to run at publish time to automatically categorize posts based on keywords moving forward.

Finally, I manually added the archive page, which lets you sort posts by category or year. This means I now have a functional archive here. Enjoy my anodyne thoughts, dear reader.

Adjacent to my day job, I’ve been toying with Claude Pro now for about a year; in my experience, it has improved significantly within the last six months. It’s not perfect: my requested edits were completed, but it also changed the CSS on the linklog so some of the text is too light to read, which I didn’t ask for and don’t want. But it has arguably extended my ability to execute on work that requires skills I don’t otherwise have (such as design and coding). What I do bring to the table is an expansive practical background in publishing and production, and all the language to describe it.