←
Steven Denney8,778 posts
Steven Denney avatar

Steven Denney

@StevenDenney86

🎓 Assistant Professor @UniLeiden. Affiliated Researcher @LeidenDH. PhD @UofT_PolSci. 👨‍🔧Techno-optimist.

◷ Joined September 2009▧ 548 Photos and videos
Andrew Piper avatar
⟳ Steven Denney retweeted
Andrew Piper@_akpiper·12h

I love the idea that non-engineers need to understand AI better and engineers need to understand culture and humanity better.

because you have not heeded my warnings and learned Spanish yet, The Point has republished in English my essay on Pangram, AI writing, the third world novel, and why so many humanists have chosen ignorance instead of grappling with what AI means for literature thepointmag.com/examined-lif…

Post attachment
Claes Bäckman avatar
⟳ Steven Denney retweeted
Claes Bäckman@ClaesBackman·20h

Researchers are finding new ways to share their work beyond the PDF: games, interactive explainers, digital editions, living papers, dataset sites. AI has made building a website almost free, and people are getting creative. Some favorites 🧵

NASA Administrator Jared Isaacman avatar
⟳ Steven Denney retweeted
NASA Administrator Jared Isaacman@NASAAdmin·18h

Voyager 1 is now one light-day from Earth, an extraordinary @NASA achievement for a spacecraft assembled by human hands and launched nearly 50 years ago. As we celebrate this milestone, with Voyager continuing her journey through interstellar space at nearly 38,000 mph, consider what those brilliant minds made possible with the technology of 1977 and how much farther we should aspire to go with the tools, resources, and ambition of our time.

Post attachmentPost attachmentPost attachment
John Burn-Murdoch avatar
⟳ Steven Denney retweeted
John Burn-Murdoch@jburnmurdoch·Oct 4

What’s the evidence on rent control? I charted the findings from 112 studies: Great for rent-controlled tenants ✅ But: Pushes up rents for other people ❌ Pushes down new housing construction ❌ Pushes down building quality ❌

Post attachment
Paul Novosad avatar
⟳ Steven Denney retweeted
Paul Novosad@paulnovosad·17h

AI is killing the old web. When an AI overview is available, Google searches result in a web site visit 40% less often. Obvs this is better for the user, but the old view-driven model of the web is in trouble.

Post attachment
Jay Van Bavel, PhD avatar
⟳ Steven Denney retweeted
Jay Van Bavel, PhD@jayvanbavel·Nov 16

What is the best AI writing detector? A study from UChicago compared Pangram, GPTzero, Originality and Roberta using multiple types of writing "Pangram maintains near-perfect accuracy across long and medium length texts. It achieves very low error rates even on shorter passages and ‘stubs." The false positive and false negative rates for Pangram are less than .05 To put this in context, this is vastly superior to any diagnost tool in mental health (eg the best measures of depression and anxiety have ~15% false positive/negative rates). bfi.uchicago.edu/wp-content/…

Post attachmentPost attachmentPost attachment
Jay Van Bavel, PhD avatar
⟳ Steven Denney retweeted
Jay Van Bavel, PhD@jayvanbavel·Nov 16

People in multiple countries consistently reject criticism of their own national group when it is attributed to a source from a different national group (intergroup criticism), as compared to when they receive the same criticism from within their ingroup. People are even willing to pay to punish outgroup critics using a measure of costly punishment (withholding lottery tickets by giving up some of their own tickets) This seems to be driven by a mix of identification (ie defending the ingroup) and social norms against criticizing outgroups. nature.com/articles/s44271-0…

Post attachment
Steven Denney avatar
Steven Denney@StevenDenney86·Nov 16

Next from our team: a companion report that puts these measures to work on Canadian data. We apply the Canaries in the Coal Mine approach (@erikbryn et al.) to the Labour Force Survey and set out what it would take for AI's effects on work to call for new policy.

Steven Denney avatar
Steven Denney@StevenDenney86·Nov 16

Indeed, one thing we missed in our review is alternative explanation on the impact side. Here, attributing the weaker labor market performance of recent grads to AI may be misplaced. WFH effects could be an important contributing factor.

New from @Stripe Economics: employment of young college grads is weaker than normal. Much of the debate is around AI's impact. But evidence is growing that work-from-home is a major factor, which we find in public data & our own Stripe data. 🧵 stripeeconomics.com/p/ai-wor… /1

Steven Denney avatar
Steven Denney@StevenDenney86·Nov 16

The largest emerging measurement gap is AI's environmental footprint. And for Canada, and others too, the biggest missing piece is individual-level data on actual AI use linked to administrative employment and earnings records.

Steven Denney avatar
Steven Denney@StevenDenney86·Nov 16

The strongest (causal) evidence concerns productivity at the worker and task level, particularly from randomized experiments. Whether these gains translate into higher productivity for firms or entire economies remains much less clear.

Post attachment
Steven Denney avatar
Steven Denney@StevenDenney86·Nov 16

AI impact is conceptually straightforward but very hard to establish empirically. Some studies point to weaker hiring of young workers in AI-exposed occupations, for example, but results are mixed across countries and measures.

Post attachment
Steven Denney avatar
Steven Denney@StevenDenney86·Nov 16

We find AI diffusion is conceptually complex but straightforward to measure. Reported adoption rates depend heavily on how AI is defined, how questions are asked, and who answers them. Workers consistently report more AI use than their employers do.

Post attachment
Steven Denney avatar
Steven Denney@StevenDenney86·Nov 16

We systematically evaluated 113 scientific sources published between ChatGPT's release (Nov 2022) and July 2026, across the six pillars of our framework. We classified each as measuring diffusion or impact and rated its research design on a five-tier evidence scale.

Post attachmentPost attachmentPost attachment