iiinova — where your digital self keeps evolving

Reposted by Cameron Blevins

Ted Underwood @tedunderwood.com · 22/09/2026

Models like Talkie-1930 sound like voices from the past. If they could reliably speak from specified historical vantage points, researchers might also use them to simulate the past. But how reliable are they? Today we release a benchmark answering that question for English contexts 1831-1930.

Free-text evaluation of answers to character modeling and constrained generation questions. This is just one of several scores Chronologic-EN can produce; we focus on it here because it's both the hardest test and the one most relevant to simulation of the past. Frontier models reach 72%; Talkie-1930 is stronger than several larger competitors, but not at the frontier by this measure. Note that this score has improved ~45% in the last two years, but still falls perceptibly short of ground truth.

516244

Cameron Blevins @cblevins.bsky.social · 09/09/2026

Thanks for this! The new history of capitalism example is really interesting. I suspect coverage of lesser known historiography would be shakier, but even seeing the shift towards nuance + citations laid out like that is useful.

120

Cameron Blevins @cblevins.bsky.social · 09/09/2026

A couple other recent posts about generative AI and teaching history that I've also found helpful: generativehistory.substack.com/p/no-more-mu... by Mark Humphries and lucaspoy.substack.com/p/the-ai-con... by @lucaspoy.bsky.social

031

Cameron Blevins @cblevins.bsky.social · 09/09/2026

I'd encourage historians to read this post from @jimclifford.bsky.social. Even if you ethically refuse to use generative AI (which I understand) I think it's important and useful to keep tabs on how quickly this tech shifts and its potential impacts on the work we do as historians.

182

Reposted by Cameron Blevins

Lauren Klein @laurenfklein.bsky.social · 09/09/2026

"Surviving the Textpocalypse" (after @mkirschenbaum.bsky.social) is the other course I'm teaching now, co-envisioned with @shsalter.bsky.social. We pushed ourselves to ask what we could really change, and how we could ever-so-slightly inflect the future. 1/ github.com/laurenfklein...

github.com

GitHub - laurenfklein/2026-textpocalypse: Class info and labs for Fall 2026 ENG 790/QTM 490: Surviving the Textpocalpyse

Class info and labs for Fall 2026 ENG 790/QTM 490: Surviving the Textpocalpyse - laurenfklein/2026-textpocalypse

3379

Cameron Blevins @cblevins.bsky.social · 01/09/2026

I would have agreed wholeheartedly with you even 12-18 months ago, but I've just seen too many examples of what these models can do now. In our field of history: generativehistory.substack.com/p/the-sugar-...

120

Cameron Blevins @cblevins.bsky.social · 01/09/2026

A colleague described this to me as seeing a shift from a bell curve distribution of student engagement to a barbell distribution. I think a core of motivated students will continue to stay engaged, but the middle might get hollowed out.

010

Cameron Blevins @cblevins.bsky.social · 31/08/2026

Thanks, I appreciate your perspective!

000

Cameron Blevins @cblevins.bsky.social · 31/08/2026

Thanks - what has worked well for you to build trust and accountability in your online asynch courses?

100

Cameron Blevins @cblevins.bsky.social · 31/08/2026

Agreed. And logistically very, very tricky when part of how we market online asynch courses is that you DON'T ever have to set foot on campus, which opens up access to a lot of students who aren't local.

110

Cameron Blevins @cblevins.bsky.social · 31/08/2026

Agreed in some ways, but online programs have vastly expanded access to higher ed for a demographic that otherwise would have been shut out.

010

Cameron Blevins @cblevins.bsky.social · 31/08/2026

As will potential students who just won't be able to take classes if online learning shutters or contracts.

010

Cameron Blevins @cblevins.bsky.social · 31/08/2026

I sometimes think we've moved the goalposts, though. Maybe your students are far more advanced than mine, but if I received this submission in 2021 I would have thought it was quite impressive.

100

Cameron Blevins @cblevins.bsky.social · 31/08/2026

Couldn't agree more, and I struggle with how to foster that engagement/culture for a lot of online classes.

110

Cameron Blevins @cblevins.bsky.social · 31/08/2026

Agreed! I started distributing a minimal note-taking template handout every day that seems to have helped a bit.

020

Cameron Blevins @cblevins.bsky.social · 31/08/2026

Love this! I've recorded the lecture portions of my classes since the pandemic, but I never have a great sense for how widely they're used by students.

010

Cameron Blevins @cblevins.bsky.social · 31/08/2026

It's absolutely exhausting - as is evaluating a pile of AI-authored assignments. Like I said, there's no good solution here.

010

Cameron Blevins @cblevins.bsky.social · 31/08/2026

Thank you for sharing, I really appreciate these! I guess my fear is that "but they can't do THIS" shrinks every year. Today's frontier models are, frankly, "smarter" than me in some ways, so it can feel hollow to make that argument to students.

100

Cameron Blevins @cblevins.bsky.social · 28/08/2026

Certainly not for multiple choice questions, but theoretically could be applied to anything with longer chunks of text involved (essay, discussion posts, etc.)

110

Cameron Blevins @cblevins.bsky.social · 28/08/2026

That's helpful perspective, thank you! And I know how much work and care goes into that kind of pedagogy.

020

Cameron Blevins @cblevins.bsky.social · 28/08/2026

Which might actually be possible at a place like MIT with their $27 billion endowment....

010

Cameron Blevins @cblevins.bsky.social · 28/08/2026

Interesting, thanks for sharing. I always tell colleagues not to treat Pangram like incontrovertible "proof" but it can be useful as one piece of evidence to consider

010

Cameron Blevins @cblevins.bsky.social · 28/08/2026

Is that allowed by your university? At least here if a class is classified as "online" you can't actually require any kind of in-person component (it would have to be "hybrid"). And we've spent the past half decade aggressively marketing all of our wonderful fully online programs.

140

Cameron Blevins @cblevins.bsky.social · 28/08/2026

It's so exhausting to keep up with individually, much less teach!

020

Cameron Blevins @cblevins.bsky.social · 28/08/2026

I appreciate all three of these. We have at least one department who has switched all of their online asynch classes to a hybrid designation so they can require students to come to campus for tests (which you can't do for an "online" designation)

040

Cameron Blevins @cblevins.bsky.social · 28/08/2026

Do you mind if I ask: how have you approached it?

000

Cameron Blevins @cblevins.bsky.social · 28/08/2026

Agreed. I keep thinking about how for either end of the student spectrum not a ton changes. Motivated students are still going to thrive in college and, arguably, might get more out of their education with AI. Some students will cheat regardless. It's the middle that I think is affected most

010

Cameron Blevins @cblevins.bsky.social · 28/08/2026

Interesting. My knee-jerk reaction has been to resist any kind of surveillance tech, but talking to a lot of faculty over the past few months has made me at least think about different options in a way I had refused to do before

140

Cameron Blevins @cblevins.bsky.social · 28/08/2026

I've had a similar experience to @tedunderwood.com in that it's largely been...fine? And at least there's an established organizational bureaucracy in place to process requests in a way that gives support to individual instructors.

140

Cameron Blevins @cblevins.bsky.social · 28/08/2026

Thanks - I really appreciated your recent post, it actually helped spur a lot of my jumbled thoughts here. lucaspoy.substack.com/p/the-ai-con...

lucaspoy.substack.com

The AI conversation should be with students, not about them

Image: Cuban literacy campaign 1960-61Thanks for reading Tomorrow Never Knows!

020

Cameron Blevins @cblevins.bsky.social · 28/08/2026

Like I said, it's not a great solution but I struggle to come up with better ones. At this point I admit that I find myself evaluating students more favorably whose prose is obviously NOT AI even if it's less clearly articulated.

110

Cameron Blevins @cblevins.bsky.social · 28/08/2026

I think about the employer piece quite a bit. So many universities are sensitive to employer demands, but I'm not sure employers even know what they want from graduates. "AI skills," currently, but I think they take for granted that those skills are built on top of a robust education.

010

Cameron Blevins @cblevins.bsky.social · 28/08/2026

I've benefited from seeing how really smart folks like @lauraknelson.bsky.social @mariaa.bsky.social, @richardjeanso.bsky.social, @tedunderwood.com, @lucaspoy.bsky.social are approaching AI and in-person teaching, and I'd (selfishly) appreciate seeing how others are approaching it in online teaching

8131

Cameron Blevins @cblevins.bsky.social · 28/08/2026

I don't like the policing dynamic of detection tools, but they seem like the least bad option here. At least they provide institutional support instead of asking over-stretched, contingent faculty to continue to bear the burden as individuals. But please: I would LOVE to hear other solutions.

8120

Cameron Blevins @cblevins.bsky.social · 28/08/2026

We desperately need solutions beyond "online asynch is toast." For years I've told colleagues that AI detection tools are unreliable and problematic. But if Pangram's accuracy results really are this robust, I think schools should think about getting a license: papers.ssrn.com/sol3/papers....

Screenshot from paper by Jabarian and Imas, "Artificial Writing and Automated Detection"

6 Conclusion
We evaluate the current batch of dominant AI-text detectors using a 1,992-passage multi-
genre corpus in conjunction with four advanced LLMs. Three key findings are obtained.
First, a clear ranking in detectors emerges. Pangram maintains near-perfect accuracy
across long and medium length texts. It achieves very low error rates even on shorter
passages and ‘stubs.’ These results are robust to exogenous changes in the threshold and the
use of ‘humanizers’ such as StealthGPT. The other two commercial detectors, Originality.AI
and GPTZero, perform well on long and medium length texts, but, depending on the
detector, struggle on short and ‘humanized’ passages. The open-source RoBERTa baseline
both misclassifies human-written texts and fails to identify up to 51% of AI-generated
passages. Second, recalculating vendor fees as cost-per-true-positive (CPTP) shows that
Pangram is substantially cheaper to implement than both OriginalityAI GPTZero. Third,
the use of policy caps facilitates cross-detector comparisons and allows the policy designer
to select a detector based on a straightforward optimization of their objective function.

2161

Cameron Blevins @cblevins.bsky.social · 28/08/2026

If you don't teach online, it's tempting to see this as someone else's problem. Putting aside the ethical arguments around accessibility, solidarity, etc. - in purely selfish terms online learning helps pay a lot of our salaries. It undergirds the enrollment and budget models for many institutions.

1182

Cameron Blevins @cblevins.bsky.social · 28/08/2026

Online classes are: a) the modality MOST vulnerable to agentic AI, b) the MOST labor-intensive in terms of faculty-to-student ratios, and c) taught by instructors with the MOST precarious employment. We need to keep this in mind when urging instructors to "teach with intentionality"

Screenshot of MIT report on AI and teaching.
2.5. Teach with intentionality
Most classes taught at MIT will require careful review and, in many cases, substantial adaptation to survive an encounter with Al without serious disruption. Al will undermine the pedagogical assumptions that guided the choice of content, the methods of instruction, and the ways of assessing student performance.
In short, it's time to reflect deeply and extensively on what we teach and how we teach it. Future class structures, assessments, and policies must be built with high intention and a clear sense of purpose, not merely tweaked in reaction to the immediate realities of Al. This does not mean, however, that every course must change in the same way.

34913

Cameron Blevins @cblevins.bsky.social · 28/08/2026

My department (History) provides a useful example: 1. Online is now 50-60% of our enrollment. 2. Online sections are, on average, nearly twice the size of in-person ones. 3. 80% of online sections over past 5 yrs have been taught by non-tenure-track faculty

1181

Cameron Blevins @cblevins.bsky.social · 28/08/2026

Implementing meaningful assessments in online courses is incredibly labor- and time-intensive. Two structural factors make the modality challenge even worse: online classes are typically bigger than in-person ones and often taught by contingent faculty.

1222

Cameron Blevins @cblevins.bsky.social · 28/08/2026

Similar result with using ChatGPT to work in Canvas to complete an unessay: it read instructor feedback on an earlier assignment, found primary sources and photos, made a zine-style pamphlet for a campus walking tour, and wrote a process reflection about "my own experience of moving through" campus

First two pages of an campus walking tour pamphlet produced by ChatGPT as part of an unessay assignment.

Screenshot of ChatGPT-authored reflection piece accompanying an unessay project:
Research and Process
I began with my own experience of moving through the Auraria Campus and realizing that its classrooms, lawns, and walkways make the former neighborhood difficult to imagine. My first idea was a general walking-tour map, but the professor's feedback pushed me to choose a specific method and a more analytical message. I decided to make a digital collage-style zine that can be read on screen or printed, folded, and carried while walking. The format combines a simplified route map, five short stops, archival material, original diagram-like illustrations, and questions that ask the reader to look closely at the present landscape.
For research, I first built a timeline from the Denver Public Library neighborhood history and the Auraria Library's displacement and scholarship page. I then used Brian Page and Eric Ross's scholarly article to understand Auraria as part of postwar urban renewal and to see community resistance as a force that shaped preservation and later politics. The 1969 Citizens for Auraria pamphlet became my main primary source because its design and wording show how supporters framed the bond as inexpensive, modern, and beneficial to everyone. I compared that public message with later estimates of the number of people and households displaced. I also used CU Denver's account of the expanded scholarship to connect the 1969 decision to a present-day institutional responsibility.

1134

Cameron Blevins @cblevins.bsky.social · 28/08/2026

Agentic AI is also kryptonite for online teaching. Here's Claude's Chrome plugin navigating a dummy Canvas shell in Student Mode to take notes on a lecture and a primary source packet, then use it complete (and ace) a quiz that I tried to ground deeply in that material.

13410

Cameron Blevins @cblevins.bsky.social · 28/08/2026

Here at CU Denver, most of our students are commuters who work full- or part-time and take a mix of in-person and online classes. About 37% of all our student credit hours are completed online and this is pretty typical across higher ed: www.insidehighered.com/news/student...

insidehighered.com

Distance Education Enrollment Surges Nationwide

From Mississippi to South Dakota, states are seeing increased enrollment in courses that rely on at least one form of technology as colleges expand online and hybrid options.

1211

Cameron Blevins @cblevins.bsky.social · 28/08/2026

Nearly all of the report's recommendations rest on an assumption that students should learn entirely in-person. This makes sense for MIT, and tbf the authors weren't trying to speak for all of higher ed. But this is simply not how most students receive a college education today.

Screenshot of MIT report on AI and education:
3.2.1. Define and communicate the value of residential
education
As a residential institution, MIT needs to clearly define the value of bringing students to campus and engaging avidly in face-to-face discussions, dialogue, and problem solving. No one reading this report was educated without these basic human interactions. But we cannot take for granted that incoming students will appreciate their value.

1262

Cameron Blevins @cblevins.bsky.social · 28/08/2026

This MIT report on AI and teaching is rightfully getting lots of praise: aiandeducation.mit.edu/report/. There's a ton to like, but it exemplifies a trend I see on Bluesky around AI + pedagogy conversations: they tend to leave out online learning entirely.

Screenshot of MIT report on AI and education:
3.1.4. Build structured in-person social learning
into subjects
Because Al tools are disrupting longstanding patterns of social learning at MIT, every subject should include a regular in-person social component (not just sitting in lecture or recitation and quietly taking notes). Instructors should intentionally structure such interactions to achieve desired learning objectives and maintain quality, even in large classes.

69637

Reposted by Cameron Blevins

Amanda Regan, Ph.D. @regan008.bsky.social · 19/08/2026

Johns Hopkins + the AHA are hosting the AI and History Conference 2026 -- Oct 15-16 in Baltimore. Whatever your take on AI (excited, skeptical, curious, already using it), there's a place for you in this conversation. Learn more at: proflouishyman.github.io/ai_conferenc...

proflouishyman.github.io

AI and History Conference 2026

A practitioner-focused conference for historians using AI and computational methods, October 15–16, 2026 at Johns Hopkins University.

047

Reposted by Cameron Blevins

Lucy Li @lucy3.bsky.social · 07/08/2026

Spent a good amount of this summer digging through a 20+ person team of teachers' free-text annotations, learning what "scaffolding" and "push for rigor" means, and iterating on this pipeline. We find that AI tutors, by default, frequently over-scaffold and rarely push for rigor.

27412

Cameron Blevins @cblevins.bsky.social · 05/08/2026

The whole thread is worth a read, demoing what it would look like for a student using Claude for Education to begin writing a research paper. This isn't learning.

020

Reposted by Cameron Blevins

Dan Cohen @dancohen.org · 14/07/2026

A new essay: “Perils of the New Armchair Scholarship” — Like nineteenth-century scholars who wrote about the world without deeply engaging with it, academics who lean too heavily on AI risk becoming paper-thin newsletter.dancohen.org/archive/peri...

newsletter.dancohen.org

Perils of the New Armchair Scholarship

Like nineteenth-century scholars who wrote about the world without deeply engaging with it, academics who lean too heavily on AI risk becoming paper-thin

052

Cameron Blevins @cblevins.bsky.social · 16/07/2026

It's been so rewarding to team up with my collaborators Eric Gonzaba, Annelise Heinz, and @regan008.bsky.social on this project and am grateful to @acls1919.bsky.social for supporting our work.

080