Jeroen Mahieu

One clear pattern that is emerging from students’ #LLM usage for their papers is that they all rely on the same theories. Like 90% uses the “resource-based view” of the firm. Before, they often relied on weird exotic frameworks, but now it has all become so boring and repetitive

On the verge of declaring defeat with chatgpt in my asynchronous online dataviz class. Something changed this semester compared to past ones and SO MANY assignments are essentially 100% LLM output.

Even the White House is using GenAI to cheat on its homework 🤷‍♂️

The State of Personal Online Security and Confidentiality | SXSW LIVE YouTube video by SXSW

Meredith Whitakker on the meaning of privacy, Signal vs Whatsapp/Telegram/iMessage/…, the value proposition of OS tech, why being a nonprofit is mission critical, and much more. Besides the privacy part this is a deep discussion on strategy in the current tech ecosystem

youtube.com/live/AyH7zoP...

Childbirth affects firm performance for female Norwegian entrepreneurs: -30 percent profit 10 years post-childbirth. No such decline for male-owned businesses is found, from John Bonney, Luigi Pistaferri, and Alessandra Voena https://www.nber.org/papers/w33448

Musk's Junta Establishes Him as Head of Government Imagining how we'd cover overseas what's happening to the U.S. right now

Having watched with growing alarm the developments of the last 24 and 36 hours in Washington, I thought I’d take a stab at how the US media would cover this story if it was happening in a foreign country. Here’s that story that should be written this weekend: www.doomsdayscenario.co/p/musk-s-jun...

Oh, the irony

All Day TA All Day TA is an AI EdTech company focused on higher education that enables professors to build customized AI teaching assistants for their courses. Available 24/7, it provides students with instant, ...

The big black box for us as teachers remains how students are actually using LLMs and how we can help them use LLMs in a way that helps them *learn*. I currently see very few efforts in this direction. Tools like alldayta.com are a first step but not the best solution for this problem imo

Because of this lack of expert knowledge, learning *from* LLMs and learning *how* to generate high-quality with LLMs is impossible. Students remain trapped in mediocre text that may *look* good but is a “sufficient” grade at best

Students lack expert knowledge which is complementary to LLM output. Without such knowledge it is very hard to direct LLMs to consistently produce output that is better than mediocre. You see this very clearly in theory sections that require careful logical argumentation and “connecting the dots”

Student effort going down. LLMs typically produce text that looks good on first sight and might trick a non-expert into thinking they can do the job with less effort from their side. However, actual quality of such first attempts is most of the time mediocre at best

However, I see little improvement among the mediocre and good proposals despite > 2 years since ChatGPT launch and model improvements. My guess this is due to different reasons:

The quality of the worst proposals has improved, mainly in terms of writing. Nobody submits terrible text anymore. My standards on this also have increased; submitting text with grammar or spelling mistakes is not done and will be penalised. 0 cost to write w/o grammar mistakes, students know this

Reading through the research proposals submitted by my students, I do not believe the claim that “student papers are dead since ChatGPT”. Despite basically 100% adoption rate of LLMs to develop their papers, there’s no sign the student paper problem is “solved”. Couple of observations:

👀

um

Does ChatGPT use 10x more energy than a standard Google search? A journey down the rabbit hole of viral AI energy claims. It's probably true in relative terms, but that's not what matters.

That’s a good article, but the figures quoted for ChatGPT vs. Google search are outdated, wrong & too high.

Best current energy estimate for a day of ChatGPT use is equiv. to driving an average car the length of a tennis court:

engineeringprompts.substack.com/p/does-chatg...

Before, I would never have even considered searching, hiring and training a TA (or two) for this project given it is too small and there is no funding. Now it costs me 10 dollars for the API and three hours to debug the code myself + some minor manual cleaning for the “special” cases. Bonkers

Eg: in less than half a day, I had GPT-4o code and run a script to extract and interpret text from unstructured scanned company pdfs in Dutch and French & return structured data based on the information from the docs (“give the gender of all the founders of the firm”). Would have taken RA weeks

An underappreciated fact about the massive cost reduction potential of LLMs for data collection and cleaning is that some projects which in the past were too costly to do compared to the possible contribution/impact are now interesting cost-benefit wise. A whole new problem space is now within reach

New randomized, controlled trial by the World Bank of students using GPT-4 as a tutor in Nigeria. Six weeks of after-school AI tutoring = 2 years of typical learning gains, outperforming 80% of other educational interventions.

And it helped all students, especially girls who were initially behind.

"I have no hopes for 2025. Humanity is disappointing. We killed the Earth. Villains triumph and the innocents suffer. I imagine these trends will continue."

incredible -- the NYT ran fluff "what i hope to see in 2025" blurbs from CEOs and economists, and then this guy

www.nytimes.com/2025/01/02/o...

To conclude: one effect being “significant” and the other “not significant” is rarely enough to conclude a meaningful difference of differences.

In fact, with a moderate correlation (most likely scenario), the difference of 0.11 on an 11-point PTV scale is probably not large enough to be statistically significant and the main claim of the paper is not valid.

For both groups the effect of inclusion is positive, and the "difference-in-differences" is 0.11 (0.20−0.09). Only with an extremely high positive correlation between VB and NVA PTV that difference will be significant. That’s not impossible, but it’s quite speculative without the underlying data.

Here's the figure with the main result. We see that inclusion/exclusion significantly altered PTV for VB (p = .045), but not for N-VA (p = .445). However, one effect being significant and the other being non-significant does not itself prove that those effects differ.

This study has gained some traction and is being picked up by the Belgian press. Looking at the results, I'm skeptical of its main claim though- that including or excluding a radical right party by the mainstream party differentially affects support for the two. Let me explain:

Opinie | De zorg-zzp’er is het symbool van een systeem dat mensen niet waardeert Zelfstandigen: Nu de overheid straks écht gaat handhaven op schijnzelfstandigheid, dreigt volgens Jeroen Mahieu een uitstroom van zzp’ers uit de zorg.

“De zorg-zzp’er is het symbool van een systeem dat mensen niet waardeert” of waarom het handhaven van de wet rond schijnzelfstandigheid zal leiden tot een uitstroom van waardevolle krachten uit de zorg. Opiniestuk in NRC.