AI in Practice on F L O T I S S E R I E

Computah: Make it a link

Thu, 13 Aug 2026 15:40:59 -0500

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.

Gender, Power and AI: Wrestling for the soul of the network, again

Thu, 14 May 2026 19:59:40 -0500

Stanford’s Clayman Institute ran a virtual panel this morning called “Gender, Power, and Artificial Intelligence,” with Safiya Noble (UCLA), Catherine D’Ignazio (MIT), Angèle Christin (Stanford), and moderator Genevieve Smith, a Clayman Institute Postdoctoral Fellow. The panel applied principles from feminist tech studies to the current moment, and covered how gender norms get encoded in data and reproduced by AI systems, and discussed whether the technology has real capacity for equitable design and implementation at scale.

Noble’s argument throughout is that the governance conversation has gotten too high-level and universalizing while the actual outputs of these systems have profound day-to-day consequences for specific people today. She named the role of AI in the recent gerrymandering of Louisiana and Indiana as examples, and called for tripling down on long-term social science research about AI’s impacts. She also pointed out that philanthropy is retreating from feminist academic and organizational work because that work originates from the same dynamics that critique philanthropy itself, precisely at a point when this research is sorely needed. A lot of money is moving in AI, and very little of it is funding the people best positioned to study how it impacts everyone downstream.

D’Ignazio was asked directly whether feminist generative AI at scale is possible. Her answer was no, with caveats, given who owns the technology today and the current emphasis on profit motive. She suggested it is more important to consider how to organize around our relationship to technology, and how we might approach questions of profit and ownership, policy and decision-making, and data and tech governance.

She provided an example of a reasonable use case by walking us through a project from her Data + Feminism Lab. The example is documented at length in her recent book “Counting Feminicide: Data Feminism in Action,” where her team partnered with activists who scour news reports to document the gender-related killing of women and girls, including cisgender and transgender women. The lab built a very lightweight AI-based approach that streamlines the scanning and identification of news stories as possible cases to include in their project, supercharging their work (note: very similar to how the NYT uses AI to analyze data for reporting). In this example, the AI’s job is task-scoped, democratically co-determined with the people who use it, and small. Smith picked this up: there is an idea baked into the current LLM moment that AI must scale to make it marketable, and the alternative is using purpose-built models that are right-sized against a body of work.

Christin spoke at length about how embodiment is one of the primary focuses of feminist theory, and how AI perpetuates the “disembodied” illusion of technology, and how this dynamic shows up in everything from the marketing to UX to user comprehension. This spoke to my thoughts on how the single-interface design of LLM chat reproduces Haraway’s “god trick,” knowledge that presents as universal while concealing the specific and situated position it comes from.

The parallel I kept returning to, listening to this, is one I think about often with my own cohort of early bloggers, women who grew up alongside the rise of the internet — and then the rise of ad tech. The internet of the late 1990s and early 2000s was being shaped by several camps: writers, students, information architects, and user-centric researchers who saw it as an information access network and a space of possibility; entrepreneurs and opportunists who saw it as a channel for marketing, monetization and extraction; and a smaller boycott camp that wanted to limit and refuse the whole personal computing and digital revolution altogether.

It was generally considered weird to be a girl on a computer or a woman on the internet — so weird that many of our peers didn’t recognize us at all — and we were there anyway, making stuff, witnessing, learning, advocating, producing, influencing. So when I watch some of my old peers, many of whom are professional writers and academics today, treat LLMs as a question of refusal rather than a condition to engage with critically, I worry we are abdicating a responsibility at precisely the moment when our technical and rhetorical expertise applies. Their refusal has good logic: user-centric researchers and communities engaged extensively with the early internet and the extractive camp won anyway, so why expect a different outcome here?

But Noble’s work on algorithmic bias attributes that failure not to engagement, but to the institutional and financial disadvantages that user-centric approaches operated under relative to gargantuan commercial interests. David and Goliath. That gap does not close through abstention. Understanding the trade-offs around tech, producing knowledge and analysis that does not depend on investors and marketers to frame the platform and the questions, requires presence. Refusal cedes so much ground.

Overall, the recommendations from the panel were practical. Noble called for people with capital (and the political will to spend it) to consider how to put money toward socially responsible research and development. D’Ignazio called for alternative funding infrastructure outside of venture capital logic, and pointed at European digital sovereignty models as worthy of consideration here. She also gestured at the popular AI Skeptics reading group as one current example of mad-and-commiserating-as-organizing that is creating safe psychological space for people to talk about AI and its tradeoffs. Christin’s recommendation was community organizing, on the grounds that LLMs are unpopular with a lot of people who feel there is no space to say so, and that finding those spaces is itself worthy because it provides shared language and awareness of others’ knowledge and experiences.

Personally, it was refreshing to hear reflections on the work (and the feelings) of being inside institutions that are being reshaped by AI, and being responsible for some of how that reshaping gets communicated and absorbed. I’m thinking about the incredible value of interdisciplinary governance, and how the commitment to governance is a specific position, and all the margins to consider.

Further reading:

Catherine D’Ignazio and Lauren Klein, Data Feminism. The foundational text on applying intersectional feminist thinking to data science practice.

Catherine D’Ignazio, Counting Feminicide: Data Feminism in Action. Extended case study of the grassroots data activism project D’Ignazio described on the panel.

D’Ignazio et al., “Feminicide and Counterdata Production.” Research paper on the counterdata methodology behind the femicide tracking project.

D’Ignazio et al., “Data Feminism for AI.” Conference paper extending the data feminism framework to questions specific to AI systems.

Safiya Noble, Algorithms of Oppression. Noble’s study of how commercial search engines reinforce racism and sexism through their ranking systems.

Donna Haraway, “Situated Knowledges: The Science Question in Feminism and the Privilege of Partial Perspective” (1988). The original essay where Haraway introduces the god trick and the case for situated, embodied knowledge against the view from nowhere.

Mon, 11 May 2026 11:16:51 -0500

Reflections on teaching fiction writing in the age of AI, from a professor with ten years of classroom experience teaching writing at MIT.

Sun, 03 May 2026 15:55:30 -0500

A new study suggests that people who use AI for writing are more able to detect AI writing than automated scanner tools. My current LLM pet peeve is how they use language like load-bearing, structural and legible to describe most ideas.

Wed, 29 Apr 2026 10:48:28 -0500

Adventures in AI: I asked a Claude agent (new Opus, Pro plan) to build a Google Doc template with multiple tabs, using an existing doc as reference. It failed three times over two days, burned thru tokens, never worked with Drive. Eventually it spat out text for me to paste into a doc I made myself.

Thu, 23 Apr 2026 17:22:52 -0500

Fellow Madisonians, someone pulled together a website ranking local businesses in Madison by how local they are (by what criteria, idk). In my experience, this is one way we’re likely to see AI used in the next couple of years, via prototyping and/or executing ideas that result in dynamic websites.

Thu, 23 Apr 2026 12:47:12 -0500

Centaurs and Cyborgs on the Jagged Frontier by Ethan Mollick in 2023: “On some tasks AI is immensely powerful, and on others it fails completely or subtly. And, unless you use AI a lot, you won’t know which is which.”

Wed, 22 Apr 2026 17:35:16 -0500

Anecdotally, I’ve seen two family court cases where one party submitted full AI chats — prompts and colorful complaints included — as formal filings. The complaints wouldn’t pass muster with a real lawyer, but the conflict was nurtured by AI nonetheless. One was dinged for wasting the judge’s time.

Wed, 22 Apr 2026 17:27:14 -0500

I’ve posted a couple of times about instances I’m aware of where people are using AI in pro se court cases, especially family courts. A new study shows evidence of increasing numbers in pro se cases at the federal level, exacerbating existing bottlenecks. Many trade-offs abound here.

Wed, 22 Apr 2026 17:10:00 -0500

A professor asked students to self-report AI usage on their homework, leading to lots of confusion and uproar. Points aside, it’s clear people want more clarity up front about when and whether to use LLM tools. In the meantime, treating students like they’re guilty until proven innocent is a bad MO.

Tue, 21 Apr 2026 10:40:52 -0500

Timothy Chester offers some thoughts on the place of AI-assisted software development in a modern research university, and suggests that just because you can doesn’t necessarily mean you should.

Tue, 14 Apr 2026 15:52:56 -0500

I’m following a guy in TX who is using AI to write and illustrate children’s books whole cloth, then self-publishes using Amazon, and getting recognition in his region as a laudable children’s author. The books are categorically not good. It’s like people are rewarding his content strategy.

Mon, 06 Apr 2026 10:05:53 -0500

Silicon sampling is the practice of using LLMs to run surveys without talking to any people at all.

Sun, 05 Apr 2026 09:21:57 -0500

Innovations in scamming: Folks are predicting that AI will supercharge scams alongside any technical and administrative innovation. Here’s one example of an unethical use of AI, where an internet-based GLP-1 hub used AI to generate fake product images and before and after photos of smiling patients.

Sat, 04 Apr 2026 10:36:44 -0500

Testing a new feature I created using a mix of open source code and Claude, hoping I didn’t break my own site. I pulled together a dynamic link library using a Hugo partial and some shortcode that automatically catalogs all of my outbound links into sortable lists.

Sun, 15 Mar 2026 08:48:15 -0500

Author Margaret Atwood plays with Claude and reports back on her experience.

Wed, 11 Mar 2026 18:37:15 -0500

Interesting read: NYT is using a custom LLM tool to track trends within the “manosphere,” as reported by the Nieman Journalism Lab.

Wed, 11 Mar 2026 18:29:45 -0500

A friend of the blog told me a story about a Substacker who uses AI to summarize books and then publishes AI-generated content about those summaries, never reading the books herself, and yet has a ton of followers. I’d guess at least some of those are purchased, betting that a high follower count will beget more followers by suggesting clout and credibility she didn’t earn as a reader talking to fellow readers. And followers aren’t subscribers, but that’s the business bet.

People are lookie-loos, they get curious when something is doing numbers and creating activity, so inflating follower counts is a real and persistent strategy. None of this is new. But best practices still hold regardless of which technologies you layer on top. Marketing erodes trust when it prioritizes short-term gains over honesty and reliability.

It’s strange to live in a time when you can’t reliably distinguish someone who has engaged with ideas from someone who automated the appearance of engaging with them.

Wed, 11 Mar 2026 13:27:52 -0500

Anecdotally hearing about LLMs being weaponized in divorce and custody, including inundating the other party with slop to drive up the opponent’s legal fees. Worse, the sycophancy is tuned to and confirms the aggrieved party’s grievances, regardless of their real-world relevance in court.

Tue, 10 Mar 2026 08:15:16 -0500

Through a new quiz, NYT asks readers to rate passages of writing against AI. Despite thinking I could spot the AI writing, my results were 50/50.

AI in practice: Chatbot tool comparison

Sun, 08 Mar 2026 13:19:11 -0500

Come look over my shoulder while I explore how and whether LLMs are good writing tools: Here’s a wee version of the LLM comparison exercise I did with my team. We’ll make it a two-fer so you can see how the “good writing” skill works in practice, though we’ll see how that actually goes.

One of the more useful things you can do with an LLM is hold up a few ideas side by side and apply lenses to them. I know this history pretty well, so I asked a series of LLMs, why is Wisconsin’s cultural identity and cohesion stronger than Indiana’s, from a historical and business perspective?

Here are the answers in one doc, for comparison.

Each LLM will give us more or less the same story, different flavor. Within the industry, the differences across the models reflect “model personality.” Asking “why” instead of “whether” will probably drive the answer to favor Wisconsin. Using multiple lenses (two states, historical + business, identity + cohesion) forces the LLM to cross-reference across more of its training data, which tends to produce a more comprehensive answer.

For all the chatter about consciousness and whatever, remember that an LLM is an infinite series of if/then/elses applied to human language and semantics, so being able to talk about language and communication, getting meta with the tool and how you think through language, helps a lot when using one. This is maybe the one thing I like about experimenting so hard with the tools. I’m thinking about the technical side of writing and enjoying it quite a lot.

Functionally: all of them acknowledge hard historical truths within the subject matter and don’t shy away from critical perspectives, which is good. Both Gemini and Copilot include in-line links, which lets you judge the output’s authority in the moment as a reader. I liked Copilot’s more than I expected here. Claude’s answers are more lyrical and do provide more context, and yet do not encourage checking against outside sources by providing links within the output. And you can see that even with the good writing skill calling out hard bans on certain structure, Claude plows right through them.

Model personality: Claude favors sociological answers to Copilot’s economic answers. Claude is also highly intellectual and narrative by comparison, and that narrative style can mask nuance by sinking relative context within the storytelling. Gemini simplifies, boosts and cheerleads where the others don’t, and really goes hard on Wisconsin’s reputation as a drinking and Packers state when there are stronger structural arguments in play. Copilot is tricky because it looks authoritative like a briefing, which also makes it easily “extractible” for the user, but every citation requires authentication unless this is one of those “good enough” tasks.

As a writer, something I find annoying across the whole spread is the semantic reveal. LLMs are semantic machines, and it is persistently revealed in ways that are weird to the human ear. All of them go out of their way to describe things as “structural,” “connective” as in “connective tissue,” “load-bearing” and “legible.”

Finally, I included a second tab where I asked Claude for analysis across the four outputs, where it suggests that my framing of the question is altogether kind of problematic. It shows how a strong prompt is sometimes also a bad approach.

There are a lot of possible takeaways here, but I’d rather set aside the question of which tool is “good” or “bad” or “better” and think more about the patterns across the tools and their implications.

Sat, 07 Mar 2026 18:48:00 -0500

One of the tricky things about consumer AI tools like Claude and Gemini is that the experience varies widely depending on the person using it, and it’s not always clear why. I have spent a lot of time learning the tools so I can advise on them in my work, and this variance of experience has become a frustrating part of the deal.

I manage a team of writers and creatives at work, and we are expected to be familiar with the tools, despite complex and sometimes hostile feelings about the political and environmental implications of this sector. That’s quite a pickle, organizationally, managerially. Borrowing from Haraway, I thought, okay, what if we take these tools seriously as a team of writers and creatives and put our professional standards up against them?

Among other exercises, I did a couple of comparisons on my team that help create discussion around the “plausibility” question. People dismiss LLM outputs as being merely plausible answers, rather than accurate or factual ones. And that’s correct; they are, and that’s the design. In many cases, plausibility is fine. Take Wikipedia, for example, which we understand to be a pretty good source, a plausible source, unless you’re writing a formal paper requiring original sources.

I digress. Ultimately, we needed to understand together that LLMs are not a WYSIWYG tool and talk through the implications.

I asked everyone to run the same paper through their LLM of choice, prompting it for a plain language summary. We then copied and pasted it into a shared doc, and compared and contrasted for discussion. Upon discussion, we had several takeaways, including that they were all similar in spirit but sometimes varying wildly in style and approach.

Knowing that algorithms are responsive and not static, we did it again later in the day and copied and pasted our outputs into the shared doc. We compared and contrasted the difference between AM and PM. Again, it was similar in spirit but varied in style and approach. Some changed dramatically. One team member whose morning summary had been jokey and conversational received a much more staid and serious version in the afternoon.

At the time, I asked Claude to explain the variance: “Even with the same prompt and source material, LLMs don’t produce identical outputs each time. This is by design — there’s a degree of randomness (called “temperature”) in how the model selects words, which means each run produces a slightly different path through the text.”

Anyway, this got our gears turning on how (and whether) to approach LLMs as a team and as individuals and led to good group discussion. (It’s important to create space for criticism and critical approaches here.) It also gave us more confidence as a team responding to this new layer of complexity in our work, and helping our professional contacts and peers think about how to approach the tools and when and whether to use them. There will be tasks where AI-based tools are “good enough,” and tasks where they are not.

The swirl of mystery and speculation around this sector has people up in arms, and it’s useful to have approaches that give people firsthand experience and to see how the experience works for others. The god trick of the singular interface turns out to be a bear for navigating it in the workplace, where our work is foundational, prosocial and specific.

Fri, 06 Mar 2026 20:31:00 -0500

Good and bad uses of AI on a website, from Sumy Designs.

Applying a Claude writing skill

Fri, 06 Mar 2026 00:47:00 -0500

LLMs have a default house writing style with identifiable patterns: sentence fragments for emphasis, “not X, but Y” constructions, lots of hard contrast, atmospheric openings, heavy use of em dashes, and heavy use of marketing language. This reflects the semantic construction of an LLM. Custom instructions can override these defaults. A custom skill is a set of instructions within your account that modify how the model generates text. When you paste instructions into your profile settings, Claude reads them at the start of every conversation and adjusts its output accordingly.

I began using Claude daily for light writing tasks about six months ago, and over that time I started cataloging the patterns I was consistently editing out, including the terrible “not X, but Y” construction that showed up in nearly every response, and persistent em dashes used as all-purpose connectors when other punctuation is more appropriate.

I went through several iterations of bullying Claude into submission, narrowing the scope each time, before arriving at this version, which focuses specifically on writing mechanics and hard prohibitions.

You’ll need a paid Claude plan (Pro, Max, Team, or Enterprise). Free-tier accounts don’t have access to custom skills.

• Within the app, navigate to Customize > Skills and Create new skills
• Select add a new skill and Write skill instructions
• Copy and paste the copy from this file into the skill, making note of the name and description boxes. Feel free to tinker.
• Save your changes.

Note: The instructions in the linked file are Claude’s work, not mine. They came out of months of conversation, where Claude would analyze my style notes, and the file evolved from there. They read a little strangely because of that process. If I’d written them from scratch, they’d sound different. But looking at the file you can see what Claude responds to and how it works.

Claude will apply these instructions to every new conversation going forward. Existing conversations won’t pick up the change, so start a fresh chat to test it. If and when Claude struggles to apply the skill, call it out specifically in the prompt, such as, “Revise this for length using the good writing skill.”

The skill specifies constraints in a few categories and the instructions are plain text. As you go, you can also ask Claude to analyze previous conversations for suggested additions to the skill, which Claude will produce and implement within the chat. Each rule operates independently, so removing one doesn’t affect the others.

Claude processes custom instructions at the start of every conversation, before it generates any output. The instructions function as constraints on the model’s default behavior. The model doesn’t always follow every instruction perfectly and the results vary by task. You will still need to edit.

Fri, 06 Mar 2026 00:19:00 -0500

WaPo on the many issues of LLM house writing styles. I have much to add. Here are some notes on applying a writing skill to override house style on Claude.

Thu, 05 Mar 2026 22:36:10 -0500

This new report from Anthropic is depressing at best, as it tries to measure which employment sectors carry the most exposure around AI expansion into the economy. In short, the tech is likely to impact two groups the hardest: educated professional women, and young workers for whom the career ladder will never materialize. In a right-side-up world, this would change the political dynamics of any policy response considerably. In this one, I don’t know.

Anthropic’s positioning here is curious, very god tricky. They are claiming the mantle of responsibility and transparency while predicting an inevitable end nobody wants, that they’re also selling as a service.

I still think much of the forecasting is oversold – the tech performs well in optimized environments, and last mile issues are a perennial concern in any engineering venture because the practical world is non-optimal. Time will tell, and there are big incentives in play. But the hunger and animus around the forecast feel bad.

Tue, 03 Mar 2026 09:59:00 -0500

Apparently one thing LLMs excel at is deanonymization at scale. The original promise of pseudonymity online was social and normative, over and above any question of technical depth: decent people don’t try to unmask you, because why. What strikes me today is how what used to be unacceptably antisocial behavior online is now both automated and unremarkable.

Over the last couple of weeks, I asked a couple of chatbots what could be known about me from this pseudonymous site, where I am more intentional about what I choose to reveal and conceal. It pulled the obvious but also drew conclusions based on a few geographic points I’d made in context that were both revealing and correct. I also noticed that it only drew from the top two pages of information - anything beyond page two of posts wasn’t part of the compute. Archives are for humans?

People assume that there is some computer magic on the backend where the LLMs connect all your account logins behind the scenes, but no, in fact it does all this through inference, by linking your digital trail, your friends, your breadcrumbs of likes and hearts and follows, and obvs your posts, into a picture of who you are, practically and demographically.

Fri, 27 Feb 2026 14:20:04 -0500

Second wavers in tech and engineering talked a lot about the “god trick” of presenting knowledge and information, particularly around math and science, in a way that suggests its objectivity is eternal, immortal, unknowable. Work by Safiya Umoja Noble and others extended this lens to Internet search and architecture, finding that the search algorithm was never neutral, instead it was a series of business decisions wearing neutrality like a costume, creating a customer service experience. LLMs take that same trick and compress it further.

It’s an old idea, and one I’ve been drawing from while I tinker with Claude, which is purportedly the best in the game. The “god trick” is baked right into the AI interface: one input, one output, an authoritative-seeming answer, offered without named perspectives behind it, trained on text produced overwhelmingly by a narrow demographic who has historically had access to both literacy and publishing, by programmers and new media drawing from the same well. Smushed together, it gives the impression that consensus exists where there are in fact many, many loose ends.

I increasingly find it annoying that even “good” AI outputs seem fixed on phrases like “key,” “core,” “exist,” “actually,” “never,” and possibly the worst sentence structure of all time, “it’s not X, it’s Y” — and I’ve begun to recognize how LLMs work like autocorrect for phrases and ideas, drawing from ranked search sources first before fanning out to more obscure sources, trying to determine and assert what’s important to me, a user known by demographics and data. It feels like a big linguistics machine, which is pretty cool in some regards, but also aggressively semantic. The math doesn’t always work to connect me to what I want to find because I am situated in my individual context in ways LLMs are not able to understand, with my memory, in my body, with my unique experiences, which shape and translate meaning for me as I interact with the world (and the web).

And so for you, in your body and memory and experience. An LLM can approximate the outputs of an experience without having access to the experience itself. Sometimes this is useful, sometimes it’s reckless.

Overall the dynamic reminds me of the famous scene from Good Will Hunting: Claude is a smart kid, and he’s never been outta Boston.

Thu, 12 Feb 2026 11:52:00 -0500

I’ve been running a series of informal experiments with my team to understand how we should advise people on using AI tools like ChatGPT, Claude and Gemini around our work.

In our last session, I gave everyone the same task: ask your AI tool to summarize a single article in plain language. Every person got a factually accurate summary, but the style and depth of analysis varied widely from person to person. We tested this further by entering the same prompt multiple times. Each time, the output was slightly different. We copied and pasted everything into a shared document so we could compare side by side.

Later that day, I asked everyone to reopen the tool they’d used that morning and repeat the prompt that had worked best: summarize this article in plain language. The outputs shifted again. Some changed dramatically. One team member whose morning summary had been light and conversational received a much more formal, serious version in the afternoon.

I then asked Claude to compare the morning and afternoon versions of one summary. “In short, the morning version emphasizes practical implications and accessibility; the afternoon version prioritizes completeness and specificity. They complement each other well — someone reading both would get a fuller picture than either one alone provides.”

Why would Claude produce these differences despite the same source material and prompt?

I asked the source.

Even with the same prompt and source material, LLMs don't produce identical outputs each time. This is by design — there's a degree of randomness (called "temperature") in how the model selects words, which means each run produces a slightly different path through the text.
But the differences here go beyond random word choices. The morning version landed on a journalistic structure while the afternoon version landed on an academic one. That likely reflects what the paper calls sensitivity to framing and order effects — one of the cognitive biases it catalogs. Small, invisible differences in how the model’s internal processing unfolds at the start of generation can cascade into meaningfully different organizational choices. Once the morning version opened with a news-style framing, the rest followed that pattern. Once the afternoon version opened with a descriptive academic tone, it followed that thread instead.

This is actually a concrete example of the robustness issues the paper identifies. The underlying reasoning about the paper’s content is consistent across both summaries — neither gets the facts wrong. But the presentation varies in ways that weren’t requested, which means someone relying on just one version would get a subtly incomplete picture depending on which run they happened to see.

It’s a useful reminder that LLM outputs are one plausible version of a response, not the definitive one.

It’s important to understand how these tools work. What looks like an authoritative answer is really just one of many possible outputs the system could have generated. The text will sound plausible whether it’s accurate or not, and you may get different results each time you ask.

Tue, 10 Feb 2026 02:26:00 -0500

I asked Claude.ai for a plain language summary of this buzzy new research paper. Claude offers the following:

Researchers from Caltech and Stanford conducted the first comprehensive survey of how and why large language models (LLMs) — AI systems like ChatGPT or Claude — fail at reasoning tasks. Rather than focusing on what AI can do well, they cataloged and organized the ways it goes wrong.
Why it matters
LLMs are increasingly used in high-stakes settings, but their reasoning failures are often unpredictable and poorly understood. The researchers argue that systematically studying failures — like accident investigations in aviation — is essential for building more reliable AI.

The researchers created a framework with two dimensions:
• Type of reasoning: informal (intuitive, social), formal (logic, math), and embodied (physical, spatial)
• Type of failure: fundamental flaws built into the architecture; domain-specific weaknesses; and robustness problems (inconsistent performance when small details change)

Informal/intuitive reasoning
LLMs exhibit human-like cognitive biases — confirmation bias, anchoring, framing effects — but without human ability to recognize and correct for them. They also struggle with “theory of mind” (understanding what others believe or intend), and with applying consistent moral or ethical reasoning.

Formal/logical reasoning
LLMs often can’t reverse simple logical relationships (if they know “A is B,” they may not infer “B is A”). They struggle to chain multiple reasoning steps together. Basic counting and arithmetic fail in ways that seem surprising given their other capabilities.

Embodied/physical reasoning
LLMs have poor intuitions about the physical world — gravity, spatial relationships, object properties — because they’ve learned only from text, not from physical experience. This extends to visual AI systems as well.

Many failures trace back to how LLMs are trained: they predict the next word in a sequence rather than reasoning deliberately. This makes them good at pattern-matching but unreliable when tasks require genuine logical inference, especially under slight variations in how a question is phrased.

Researchers have proposed fixes including better training data, techniques that force step-by-step reasoning (like “chain-of-thought” prompting), connecting LLMs to external tools like calculators or physics simulators, and architectural changes. However, no single fix is comprehensive — many improvements in one area don’t transfer to others.

Sun, 08 Feb 2026 13:43:00 -0500

Something else I noticed in my experimentation with AI and creative writing: Claude prefers a mid-length, declarative sentence, while I prefer a lot of variety in my prose. Sentence variety is a primary consideration in any text-based communication approach. Write accordingly.

Sun, 08 Feb 2026 12:58:00 -0500

I have a confession. While experimenting with AI over the last year, I wondered what would happen if I crammed an unfinished novel draft, one I actually care about, into Claude. Claude is pitched as the LLM for writers, with Claude 3.7 Sonnet and 3 Opus widely regarded as the premier LLMs for writers, including creative writing, long-form content and human-like prose. Meanwhile, I majored in English and work in mass communications, so I’m trained to think about writing creatively, strategically and tactically. Writing and personal expression have been part of my daily life for most of my life. If this tool could in fact produce a quality story, someone like me should be able to make it happen. Instead, the experience left me confident that AI isn’t a good vehicle for creative, narrative writing.

Here’s what I found:

On the technical side, Claude struggled to maintain a narrative thread over time. The longer the chat, the more the bot drifted and eventually lost track of details and claims made about characters earlier in the plotline. It’s not a sustainable approach for narrative writers because continuity matters: outsource too much plotline to the bot and your characters lose relationship to one another.

LLMs like Claude work fine for writing support—they can function something like a synonym machine, helping writers work through technical questions of redundancy, register, length, and other semantic needs while drafting. But when you outsource world-building and meaning-making to an LLM, it becomes narratively confusing fast. Despite giving Claude extensive background on my primary characters and the world they live in, it would confidently declare that a character’s relationship to another was X, then claim the opposite on the next page. Dialogue was thin and expository. It preferred a sort of “maid and butler” style of dialogue where two characters artificially recap shared knowledge for the reader. Meanwhile Claude does not do feelings well, which is arguably the point of much narrative writing.

Ultimately my drafts were worse off than what I started with – less organized, more confusing, with so much narrative drift that almost nothing was usable, even as a first draft. A devil’s advocate might argue that my prompting wasn’t sophisticated enough to produce the results I wanted. Sure.

But then we have the second problem: Claude’s approach to storytelling isn’t narratively interesting. Fiction and narrative writers put tremendous energy into world-building and sensory experiences. The goal is to immerse the reader in a sensory experience so total that they can experience another world entirely – the original VR, if you will. A great writer even exploits your higher-level cognitive functions by reusing parts of the brain that evolved for action and perception, which is why a good story makes you think, feel, and wonder.

Claude does not feel or wonder. Claude collates.

A key part of this essay suggests that LLMs create meaning through triangulation – that by pinging other ideas and vocabulary, an LLM can get a human reader close, or close enough, to suffice in many cases of writing. In my experience, this is true enough in business writing, where tinkering with approach and register can become as important as precise verbiage.

But this misses the pleasure and the point of good storytelling, which is myriad but usually centers on the satisfaction of expanding your imagination and experience through narrative, by seeing your own messy, striving, failing, hopeful, and collective human experience reflected in another person’s expression. That kind of meaning-making doesn’t happen through triangulation. It happens through the labor of human thought, experience and skilled articulation. That’s art, babes.

This article gets into the mess of AI and creative writing, within the domain of the romance genre, which famously cranks out variations on romance themes at a rapid clip. It drills down into some of the debates about writing, authority and authorship in relationship to LLMs that are playing out across the publishing sector now. Remember: early research suggests that most writers who use LLMs as part of their workflow ultimately retain their sense of authorship in and around the tools, suggesting that even when writers adopt AI assistance, they still see themselves, not the tool, as the creative and accountable source. So based in my experience above, I suspect that if an AI approach to creative writing is successful, it’s because the author is linking her approach to emerging tech, not because the work is good, and that’s a difference worth distinction.

Thu, 05 Feb 2026 23:03:00 -0500

The reporter who tried to replace herself with a bot

Crunching for clarity

Fri, 30 Jan 2026 11:12:00 -0500

In 1999, academic and theorist Judith Butler famously won an award for the worst academic sentence, raising good questions about how we read difficult texts, who gets to access academic ideas, and the role of academic and plain language in and around the academy:

The move from a structuralist account in which capital is understood to structure social relations in relatively homologous ways to a view of hegemony in which power relations are subject to repetition, convergence, and rearticulation brought the question of temporality into the thinking of structure, and marked a shift from a form of Althusserian theory that takes structural totalities as theoretical objects to one in which the insights into the contingent possibility of structure inaugurate a renewed conception of hegemony as bound up with the contingent sites and strategies of the rearticulation of power.

This was my era in college, and I remember feeling challenged and overwhelmed as an undergrad trying to read theory for the first time. So today, I asked Claude.ai to put it in plain language, just to see what would happen. Here’s the after:

Scholars used to think that capitalism shaped society in consistent, predictable patterns. But newer thinking focuses on how power works—how it repeats, overlaps, and gets reshaped over time. This brought time into the picture and moved away from older theories that treated social structures as fixed. Instead, recognizing that structures can change opened up a new way of understanding power: it's tied to specific moments and strategies, and it's always being renegotiated.

A meta lesson about AI assistance

Sun, 07 Dec 2025 15:01:00 -0500

I just completed my first attempt at coding using AI, in this case having Claude assist me with putting together a simple client-side OPML parser using Dave Winer’s Feedland service.

Winer’s original script is pretty slick, and includes a list of all my feeds with titles, URLs, and categories; click-to-expand functionality to see the 5 most recent posts from each feed; clickable post titles that open articles in new tabs; sort options (by title or by update); and automatic updates when I change my FeedLand subscriptions.

You can check it out here: Feeds

The official documentation method didn’t initially work because Hugo (the blogging software behind micro.blog) was wrapping client-side templates around the script. The toolkit requires server-side dependencies that don’t exist on static sites like micro.blog, and we hit a cascade of missing JavaScript dependencies (jsonStringify, servercall, etc.). Each fix revealed another dependency, leading to some “sunk cost” frustrations for me. I kept trying because I wanted to see if Claude could pull it together. Through trial and error, I got to a point where the OPML file was rendered correctly without server dependencies or complex external libraries.

Time invested: ~3 hours (including wrong turns)

Time it should take: 10 minutes

AI extended my code reach beyond my practical skillset by quite a lot. I now have a dynamic and dedicated place to read and share news feeds as I wish. Though even when generative AI works and works well, I have significant concerns about the intellectual property implications of AI, and this project brought those tensions into sharp focus. The AI could only help me because it was trained on documentation and intellectual work from the open source community, contributions made freely in the spirit of knowledge sharing, not to train commercial AI systems. I tapped into their expertise by paying Anthropic $15 a month. While I’m grateful for the accessibility this provides to non-developers like me, I recognize there’s an unresolved ethical question about whether this use respects the intent and labor of the original creators. The feat is incredible; the foundation it’s built on deserves careful consideration.

After the exercise was complete, I asked Claude how I could have improved my prompting to make this process easier, and in short, Claude said I could have been a web developer. But since I’m not, here’s what it recommended:

✅ When the process isn’t working, question the process mid-stream. Most people either give up or keep following bad advice deeper into rabbit holes. Stop and question the LLM’s process and ask for alternatives to force a reset.

✅ Push for usability. Keep bringing the conversation back to what you actually need the end result to do, not what’s technically impressive or “correct.” In my case, this meant repeatedly asking “can I click through to the articles?” rather than getting lost in discussions about CORS proxies or JavaScript syntax. Focus on outcomes, not implementation details.

✅ Ask for complete solutions. Instead of trying to mentally patch together incremental changes across multiple responses, ask the LLM to provide fresh, complete code each time. This prevents copy-paste errors and ensures you’re always working with a coherent, tested solution. There’s more than one way to crack an egg, but you want the whole egg regardless.

After all that, I got it to work but can’t figure out how to make it show up in my header menu, with or without Claude. TBD.

Mon, 17 Nov 2025 09:40:59 -0500

Human in the loop (HITL): HITL means that humans are involved at some point in the AI workflow to ensure accuracy, safety, accountability or ethical decision-making. HITL inserts human insight into the “loop,” the continuous cycle of interaction and feedback between AI systems and humans.

⚡ AI's last mile problem in higher ed

Sun, 17 Aug 2025 09:15:00 -0500

I graduated from college right before the 2008 recession and bounced through some unpromising temp jobs until an opportunity emerged for a permanent position. Sometimes you just need to get in where you fit in, and so I did. That’s how I came to work for a regional cable company that used federal money to expand the new national broadband network, extending out to the rural communities dotting central Indiana.

It was a front row seat to the national broadband expansion efforts of the early 2000s. Our business ran right across the state, spanning the 80 or so miles from Attica to Kokomo, which included several small cities with large manufacturers, two public research universities, and several community and liberal arts colleges. The strip of broadband fiber at the core of our service followed existing highways and electrical lines that split the corn and soybean fields from town to town, feathering out to more rural areas from there.

I worked a variety of roles there that put me face-to-face with a classic technical problem: the last mile. On many occasions, someone would come in looking pensive, and explain that the fiber had been extended all the way from town to their hamlet, and yet there was no plan to connect their property to the pole. Over time the pattern was clear: while the network was expanded, the cost of running a physical line to each individual property was too expensive and specific at scale. These customers often left without a path forward despite all their efforts and ours.

The “last mile problem” refers to the logistical challenges and high costs associated with the final leg of delivering goods or services to the end customer. It’s often the most difficult and expensive part of the supply chain, despite being a relatively short distance. The pattern shows up everywhere: public transit can get commuters most of the way most of the time, but that final leg of the journey remains specific and individual and problematic. E-commerce companies promise drone delivery solutions, scooter and bike-share apps claim to solve urban mobility gaps, but these technological optimizations remain persistently stubborn at scale, running up against the messy realities of sidewalks, intersections, and actual human behavior.

Tl;dr: I’ve been turning over this suspicion that AI automation will hit a classic “last mile problem,” especially in the public sector.

AI systems, particularly LLMs, are like those systems—they work incredibly well in their intended domain, processing and manipulating information. But because they’re fundamentally an information-only approach, that creates their own last mile problem when we try to implement them in physical and context-specific environments. Public institutions are uniquely specific — they are often the originators and producers of knowledge and the keepers of original policy, tasked with making the rubber hit the road. Additionally, the need for comprehensive data protection required by public workers and institutions fundamentally hamstring potential applications.

Actually implementing recommendations is where you hit the last mile. This is the work of public administration.

Imagine an AI application trained on every facilities management manual ever written and tuned to synthesize best practices for HVAC optimization. It can analyze years of energy usage data and recommend precise temperature adjustments for different zones of a building, but can’t feel that the third floor is always stuffy, or know that the facilities manager retired last year and took decades of institutional knowledge with him. Building A’s HVAC system was installed in 1987 and has a manual keypad. Building B’s system interfaces with the campus-wide monitoring system, but unreliably, and investigation is slated for later, someday, when resources allow. Professor Smith has taught in Room 204 for 25 years and will blow up your spot before moving to a different classroom for maintenance. Your engineers who manage these spaces are balancing human teams, who have time off and training and other priorities they manage in life. So, you need staff who understand the quirks of each area, the history of each system, the politics of which departments will accept changes and which will flood your inbox with complaints. You need someone who knows that the third floor always runs hot because of a design flaw from 1974, and that the solution isn’t more precise control but a $50,000 renovation that’s been deferred for a decade because a glittering new project across campus takes priority.

Imagine this tangle of questions and contingencies times infinity on every university campus in existence. Universities are like cities—they’ve been built and rebuilt over decades or centuries, with layers of systems and fiefdoms that weren’t designed to work together. AI recommendations assume a level of standardization that simply doesn’t exist. Every AI implementation in higher ed requires navigating multiple constituencies with different priorities and power structures. It’s like trying to redesign traffic patterns in a neighborhood where the residents, business owners, commuters, and city planners all have veto power and conflicting interests.

So. When looking at efficiency efforts spinning up across the education sector, I’m feeling pensive, trying to understand how exactly the house gets connected to the pole.

The promise of new tech in higher ed needs to more deeply consider the translation costs: the human labor, institutional knowledge, knowledge documentation and local adaptation required to bridge between the usefulness of tech and specific realities of public university work. Public employees want modernization and don’t want to fall behind. We want systems that work. We are also balancing a great deal of change and pressure as a sector, with fewer material resources than ever. We need less marketing and more right-sizing in the claims around AI against the political and tech realities of public administration.

This disconnect between technological promise and implementation reality becomes even more critical as higher education faces increased political scrutiny. When tech vendors promise that AI will solve efficiency problems or reduce administrative costs, institutions are under immense pressure to deliver measurable results quickly. But the translation costs we experience don’t disappear just because the political pressure to modernize increases.

The institutions that thread this needle will be the ones that accurately assess these translation costs upfront and set expectations accordingly—not the ones that assume the technology will magically bridge the gap between digital and physical, abstract and specific.

⚡ Navigating AI in higher ed communications: A practitioner's guide

Fri, 15 Aug 2025 11:49:12 -0500

by Lauren Bruce

As communications professionals in higher education, we work for institutions built on the pursuit of knowledge and innovation, yet many of us feel uncertain about how to thoughtfully integrate one of the most significant technological advances of our time: artificial intelligence.

Over the past year, my team has wrestled with questions that didn’t exist in our profession just a few years ago. Should we use AI to draft articles and email copy? How do we disclose AI-generated content, or do we? When does AI assistance cross the line from helpful tool to ethical concern?

These aren’t abstract questions any longer. Over the last year, I had to overcome AI resistance of my own to develop practical, hands-on approaches to AI use that align with our institutional values while acknowledging the realities of modern communications work (more on that below). What I’ve learned is that the answers aren’t found in blanket policies or rules, but in applying our existing professional ethics to these new tools. Here is where I am today on the journey from AI praxis to practice.

Mission first

The foundation of responsible AI use in our field starts with a principle we already know: everything we do should advance our institution’s educational mission. Higher education exists to create, share, and preserve knowledge while fostering critical thinking and diverse perspectives, in service of students, faculty, researchers, workers and the world.

The bulk of our work comes from conversations with colleagues, understanding of our campus dynamics and processes, and professional judgment about what our community needs to hear. This inevitably means more work upfront, but it maintains the authenticity and institutional knowledge that our audience deserves, regardless of whether AI tools are part of the process.

Transparency without paranoia

Do I need to mention AI every time I use it? The answer isn’t simple, but I’ve found a helpful framework: consider whether your audience would feel misled if they knew how AI was involved in creating the content.

When I use AI to polish grammar and shape format, that feels similar to using spell-check – it’s helpful but not something that changes the fundamental nature of the content we wish to communicate. But when AI helps generate the main structure for a story about campus policy changes, that’s a different ball game. The audience expects those priorities and framing decisions to come from human judgment about what matters to our community.

Internally, we differentiate between the two by defining whether or not you are “automating” processes using AI, or “augmenting” processes using AI. Full disclosure, my area of experience is in augmentation, not automation. That said.

I’ve started recommending simple disclosures when AI plays a substantial role in content creation. A line like “This article was developed with AI assistance” maintains trust while allowing us to thoughtfully benefit from these tools. It’s not about being defensive, it’s about being transparent with the people we serve, especially as the tech and attitudes around it evolve over time. Additional qualifications can be included here, such as how the information was shaped and shared by AI or not (privacy implications abound).

Here, it’s important to remember to only use your university-approved tools, because university enterprise AI tools are modified to meet campus rules and requirements related to data handling.

The accuracy imperative

Perhaps nowhere are the stakes higher than with accuracy. In higher education communications, we’re not just sharing information—we’re stewarding public trust in our institutions and, by extension, in higher education itself. In addition, much of the information we are communicating is original, in that it’s new information that cannot be generated using the limitless soup of generative AI.

Every piece of AI-generated content requires human verification, especially anything involving numbers, research findings, or claims about institutional achievements. This means checking sources, confirming statistics, and ensuring that quotes are accurate and properly sourced. It’s more work, but the alternative—publishing incorrect information—could undermine years of relationship-building with community stakeholders and partners.

The promise of speed and efficiency that comes with generative AI must be balanced with the work of close reading, the skill and practice of carefully analyzing a passage’s language, content, structure, and patterns in order to understand what a passage means, what it suggests, and how it connects to our larger body of work. I firmly believe that close reading, learned in the Humanities and Social Sciences, will become increasingly important to understand, shape and steer AI output, especially with regard to public communication best practices.

Inclusion as a practice

AI bias isn’t an abstract concern—it shows up in subtle but significant ways in the work. I’ve noticed that AI tools often default to formal, academic language that might exclude first-generation college students, or suggest examples and metaphors that assume certain cultural backgrounds, for example.

This has made me more intentional about prompt engineering—the way I request AI assistance. I build digital accessibility and plain language best practices into my prompts, in alignment with institutional best practices. One tip is to draft the original using my chosen, intentional language, then ask for revisions using as much of the original verbiage as possible. The difference in output is significant and it allows me to focus on higher-order communication strategy while demonstrating both accuracy and inclusive values in our output.

Privacy and the long view

Working at a public university means balancing transparency with appropriate privacy protections. We work within strict guidelines about what information can be included in AI prompts, particularly around student data, personnel information, and strategic planning discussions. Again, it’s important to only use your university-approved tools, because university enterprise AI tools are modified to meet campus rules and requirements related to data handling.

The challenge is that AI tools work best with context, but providing that context can sometimes mean sharing information inappropriately. I’ve learned to be creative about how I frame requests to AI tools—giving enough context for useful output while protecting sensitive information about individuals and institutional operations.

I focus AI prompts on publicly available information rather than including details from internal planning discussions or individual faculty concerns. It requires more thoughtful preparation, but it ensures we’re protecting appropriate confidentiality.

Speed vs. strategy

The efficiency of AI is seductive, especially when facing tight deadlines and endless communication requests. But I’ve learned that speed can’t come at the expense of quality or authenticity.

Authentic institutional voice and authority doesn’t emerge from algorithms—it requires the deliberate application of human judgment to ensure our plans and communications reflect our campus culture, embody our values, and resonate with our specific audiences. The strategic thinking we bring—our ability to read context, navigate relationships, and understand the subtle dynamics of higher education communication—cannot be automated.

Consider my own practice: I frequently engage AI as a collaborative thinking tool, particularly for structural planning and format development. However, AI’s default tendency toward comprehensive, multi-layered approaches often produces unnecessarily complex frameworks for university communication realities. This is where professional judgement becomes critical. Strong strategic foundations and institutional knowledge allow us to right-size AI’s expansive suggestions into focused, contextually appropriate communication plans that actually serve our goals and communities.

Looking ahead

What I’ve learned over this past year is that responsible AI use isn’t about following a rigid set of rules. It’s about applying the professional ethics we already have to new technological capabilities. The core principles that guide good communications work—accuracy, transparency, service to mission, respect for audience—remain the same.

What’s different is that we now have tools that can enhance our ability to live up to those principles, if we use them thoughtfully. AI can help us communicate more clearly, research more efficiently, and reach broader audiences. But only if we maintain our professional judgment about when, how, and why to use these tools.

As our field continues to evolve, I’m convinced that the communications professionals who thrive will be those who can harness the power of AI while maintaining the human insight, ethical judgment, and institutional knowledge that define excellence in our profession. The technology will keep changing, but our commitment to serving our institutions and communities through ethical, effective communication remains constant. That’s the foundation we build on, whether we’re writing with pen and paper, collaborating in a digital document, or prompting the most sophisticated AI tool on campus.

*This post reflects my ongoing learning about AI ethics in communications practice and was generated using the assistance of AI (Claude, Gemini). Cross-posted on LinkedIn.