iiinova — where your digital self keeps evolving

Reposted by Gideon Moore

Riley Acton @riacton.bsky.social · 14/09/2026

October is shaping up to be Econ of Edtober in the @bowdoincollege.bsky.social econ department! So looking forward to both of these visits. @gideonmoore.bsky.social @joshua-goodman.com

0142

Reposted by Gideon Moore

Sam Thau @samthau.bsky.social · 01/09/2026

In a new @siepr.bsky.social WP, @gideonmoore.bsky.social, @jdlight.bsky.social and I document a small but broad shift towards more political, more liberal courses across nearly 1,000 US colleges 2000-present. We then study how students + faculty shape political content. A thread (1/N)...

$Abstract: We measure two dimensions of politics in college courses --- the amount of political content and its partisan orientation --- using course descriptions from nearly 1,000 U.S. colleges and universities over the past 25 years. Using a novel text-as-data method, we separate the amount of ``political content,'' the extent to which a course engages with topics salient to politics and public policy, from ``slant,'' the partisan direction of that content. Since 2000, the average course has 0.15 SD more political content and 0.13 SD more liberal slant. Changes are most pronounced in the tails of the distribution, particularly for slant, and are modest relative to persistent cross-field differences. Following instructors who move between institutions, we find that instructors account for 60-65\% of cross-sectional differences in course political content and slant. Using enrollment data, we find students show no preference for liberal content before the 2010s; demand for liberal content rises over the following decade, peaks in 2020, and falls slightly thereafter.$

154

Gideon Moore @gideonmoore.bsky.social · 05/08/2026

Excited to have this out! Find our discussion of the paper here: bsky.app/profile/did:...

010

Gideon Moore @gideonmoore.bsky.social · 05/08/2026

Bottom line: profile data is powerful, but it's more equilibrium object than archive. Full paper linked below. Thanks to @reveliolabs.com for sharing the data and @stanfordimpactlabs.bsky.social for funding this project. END. (16/16) bsky.app/profile/nber...

000

Gideon Moore @gideonmoore.bsky.social · 05/08/2026

Digging more into the MBA result, we see LLM use is negatively correlated with program prestige. M7 program grads are half as likely to use LLMs to update their profiles relative to the average MBA. (15/16)

100

Gideon Moore @gideonmoore.bsky.social · 05/08/2026

LLM use (orange) is heavily stratified by education---but opposite the gradient found by Deming et al. High school grads use LLMs the most, PhDs the least. MBAs use them significantly more than other Master's holders. (14/16)

100

Gideon Moore @gideonmoore.bsky.social · 05/08/2026

Finally, we can measure AI adoption for this task by looking for LLM-like language. Building on work from @hippopedoid.bsky.social et al., we search updates for words like "delve" that LLMs use at much higher rates than humans, letting us identify which groups lean on LLMs the most. (13/16)

110

Gideon Moore @gideonmoore.bsky.social · 05/08/2026

DEI: In the first half of our sample people regularly added DEI references to their job titles. However, we see a large purge at the beginning of 2025 coinciding with executive orders 14151 and 14173 against DEI. (12/16)

110

Gideon Moore @gideonmoore.bsky.social · 05/08/2026

WFH: Directly after the pandemic, workers regularly added work-from-home references to their job titles. However, this has declined steadily since then---with WFH being net *removed* in recent months. (11/16)

100

Gideon Moore @gideonmoore.bsky.social · 05/08/2026

AI: Starting in early 2023 after the launch of ChatGPT, there has been an explosion in workers retroactively adding AI to their job titles. At its peak, this reached more than 5x baseline levels. (10/16)

100

Gideon Moore @gideonmoore.bsky.social · 05/08/2026

By examining what workers change when time traveling, we can capture what they believe firms want---that is, second-order beliefs about skill demand. As proofs-of-concept, we track three skill sets using title updates. (9/16)

100

Gideon Moore @gideonmoore.bsky.social · 05/08/2026

How do we know that workers update their profiles to improve their job search? Because workers who change jobs time travel at nearly twice their base rate just before switching, then return toward base level. (8/16)

101

Gideon Moore @gideonmoore.bsky.social · 05/08/2026

Papers using these data often restrict to highly-connected profiles to enforce a minimum level of quality. However, this makes the time travel problem worse, not better: the most connected profiles are nearly 3x more likely to time travel than the least connected profiles. (7/16)

100

Gideon Moore @gideonmoore.bsky.social · 05/08/2026

With 1 in 5 users time traveling, this phenomenon is widespread. However, it is unevenly distributed: the rate is 1 in 3 in tech, but only 1 in 7 in real estate. This yields the expected geographic distribution: time travel is 50% more common in the Bay Area than in Houston or Philadelphia. (6/16)

110

Gideon Moore @gideonmoore.bsky.social · 05/08/2026

Of particular interest are "time travelers:" people who update descriptions of jobs they have already left. The job is in the past, so these adjustments can't reflect real changes in responsibilities---only changes in how workers *advertise* their old job. (5/16)

100

Gideon Moore @gideonmoore.bsky.social · 05/08/2026

Using monthly vintages of Revelio Labs data, we can "roll back the clock" to compare LinkedIn today vs. historically. Our final panel contains monthly observations of nearly 30 million active American profiles 2021-2026. (4/16)

100

Gideon Moore @gideonmoore.bsky.social · 05/08/2026

This leads researchers to overestimate historic supply of skills desirable in today's economy. For example, measuring the quantity of AI skills pre-ChatGPT based on LinkedIn today (orange line) overshoots the true record (green line) by around 30%. (3/16)

110

Gideon Moore @gideonmoore.bsky.social · 05/08/2026

Researchers rely on LinkedIn profiles to understand workers' employment histories. However, workers update these documents as part of the job search process. As they polish their resumés, their job descriptions become less like the job they had and more like the job they're looking for. (2/16)

100

Gideon Moore @gideonmoore.bsky.social · 05/08/2026

1 in 5 US LinkedIn users edit job descriptions on well after leaving the position. In a new WP with @nickbloom.bsky.social, Lisa Simon, & Caelan Wilkie-Rogers, we argue this contaminates work using profiles as employment histories-but enables us to track skill demand over time. 🧵 (1/16)

172

Reposted by Gideon Moore

SIEPR @siepr.bsky.social · 03/08/2026

🆕 Study looks at changes job seekers make to their LinkedIn profiles to match perceived demand for skills. AI mentions are ⬆️, DEI is ⬇️. SIEPR's @nickbloom.bsky.social, @gideonmoore.bsky.social & economists at @reveliolabs.com also identify heavy users of AI for resume help: tinyurl.com/3fvhne5d

031

Reposted by Gideon Moore

NBER @nber.org · 06/06/2026

Examining how much fraud there is in Medicare and who commits it, from Liran Einav, Amy Finkelstein, Yunan Ji, @nealemahoney.bsky.social, and @gideonmoore.bsky.social www.nber.org/papers/w35280

061

Reposted by Gideon Moore

Adam Serwer @adamserwer.bsky.social · 05/12/2024

Wrote about what’s behind conservatives watching movies and shows warning about the dangers of fascism and somehow coming away thinking the fascists are the good guys, and they should follow their example www.theatlantic.com/ideas/archiv...

theatlantic.com

Republicans Are Confused About the Concept of Bad Guys

Some conservatives are embracing the villains in what are supposed to be cautionary tales.

21543301158