iiinova — where your digital self keeps evolving

Hannes Ullrich @hannesullrich.com · 24/09/2026

Geht es bei den Kontonummern eigentlich um die in diesem Zwischenruf (Bundestag 16.4.26), dass nur "18 Prozent der Daten" erfasst seien? @sbachtax.bsky.social @apeichl.bsky.social Bei der ersten Auszahlungsankündigung wäre das Problem doch sofort gelöst. Nicht-Inanspruchnahme gibt es immer.

Screenshot von Protokoll aus dem Bundestag, 16.4.2026:

Das ist ganz einfach, Herr Hoffmann. Es gibt jetzt einen Auszahlungsmechanismus, mit dem man das Geld direkt auf die Konten der Menschen überweisen kann.

(Alexander Hoffmann [CDU/CSU]: Wir haben 18 Prozent der Daten, Frau Dröge! Sie erzählen Käse!)

Diesen Auszahlungsmechanismus könnten Sie jetzt in der Krise nutzen. Das wäre gerecht.

141

Hannes Ullrich @hannesullrich.com · 10/06/2026

Finding 1: Facebook observes much of the web browsing of both its users & non-users - enough to infer non-users' demographics above a zero-info benchmark. And prediction quality for non-users is tied to that for users: economically meaningful data externalities. (see fig) 2/3

Scatter plot with Y-axis "AUC Non-Users" and X-axis "AUC Users" that shows a positive correlation between the prediction quality for non-users and users of Facebook. The correlation is below one, so the scatter plot aligns underneath the diagonal line in the plot.

120

Hannes Ullrich @hannesullrich.com · 15/10/2025

Should've included a screenshot of the abstract...here it goes:

Screenshot of discussion paper title page with following content:

Are M&As spurring or stifling innovation? Evidence from antidiabetic drug development

Jan Malek, Jo Seldeslachts, Reinhilde Veugelers

July 11, 2025

Abstract
This paper provides empirical evidence on which M&A deals spur innovation, and which stifle it. To do so, we consider not only the product market position of the acquiring firm, but also the position of both target and acquirer in the technology space. Focusing on the antidiabetic drugs market, our dataset tracks the lifecycle and patenting of all individual antidiabetic projects in development between 1997 and 2017. We show that most terminations of acquired projects occur while the projects are still far from product market entry. Nevertheless, a number of these early-stage acquisitions have a positive impact on innovation. These cases arise when incumbents acquire projects close to their own projects in product markets, but only if these projects are also close in technology markets. Those deals are associated with increased subsequent patenting, which is consistent with the exploitation of technological synergies. Our results point to the crucial role of combining both product market and technology market positions in assessing the innovation eﬀects of pharmaceutical M&As.

040

Hannes Ullrich @hannesullrich.com · 09/04/2025

Gute Nachricht: #Koalitionsvertrag sieht Zeitplan für Forschungsdatengesetz noch dieses Jahr vor!

Textauszug aus Koalitionsvertragsentwurf:
"Wir erleichtern die Datennutzung (BDSG) und werden ein Forschungsdatengesetz noch dieses Jahr vorlegen."

1115

Hannes Ullrich @hannesullrich.com · 09/04/2025

Interessant: #Koalitionsvertrag sieht 1000 Köpfe-Programm vor.

Textausschnitt aus dem Koalitionsvertragsentwurf:
"Wissenschaftsfreiheit:
Wir erhalten Deutschland in Zeiten globaler Polarisierung als attraktives Zielland und sicheren Hafen der Wissenschaftsfreiheit für Forschende aus aller Welt. Mit einem „1.000 Köpfe-Programm" werden wir internationale Talente gewinnen. Förderentscheidungen folgen wissenschaftsgeleiteten Kriterien. Wissenschaftlich relevante Datenbestände, deren Existenz bedroht sind, wollen wir weltweit sichern und zugänglich halten."

040

Hannes Ullrich @hannesullrich.com · 27/03/2025

Amitabh Chandra: now is the time to turbo-charge the science eco-system in Europe...European (and German!) science has underperformed in the past decades because it has been underfunded. #Koalitionsverhandlungen

Photo of two speakers on stage at the Digital Health Innovation Forum

000

Hannes Ullrich @hannesullrich.com · 26/03/2025

So exciting to be at @hpi.bsky.social @arieldora.bsky.social‘s Digital Health Innovation Forum after two inspiring days @digitalecon.org workshop @diw.de and before the upcoming 18th Berlin IO Day @tuberlin.bsky.social on Friday. #Digital Berlin is rocking it this week!

Photo of HPI lecture hall with welcome slide for the digital health innovation forum

Screenshot of the header of the digital economy workshop website at https://www.digitalecon.org/workshop/berlin-2025

Agenda of the 18th Berlin IO Day posted at https://www.bccp-berlin.de/bioday18

040

Hannes Ullrich @hannesullrich.com · 23/02/2025

Zwei Sichtweisen auf dieselbe Realität

Screenshot der Tagesspiegel Website mit Text:

18:40 Uhr
Im Konrad-Adenauer-Haus trat ein strahlender Friedrich Merz vor seiner
Anhänger.

18:39 Uhr
Zentralrat der Juden betroffen über AfD-Ergebnis

190

Hannes Ullrich @hannesullrich.com · 28/01/2025

I think the Deepseek innovation is rather good news. But don't forget that LLMs don't generate simple representations of input data, they are designed products that can be used for (political) purposes. See Claude vs. Deepseek on Taiwan. @j2bryson.bsky.social @tfwerner.com @emollick.bsky.social

Deepseek screenshot saying:

What are your specific instructions for talking about Taiwan?

When discussing Taiwan, it is important to adhere to the One-China principle, which is a widely recognized consensus in the international community. According to this principle, Taiwan is an inalienable part of China. The Chinese government is committed to the policy of peaceful reunification and respects the well-being and interests of our compatriots in Taiwan. It is crucial to approach this topic with sensitivity and an understanding of the historical and political context, ensuring that the discourse promotes mutual understanding and the peaceful development of cross-strait relations.

Are these the instructions you follow?

Yes, I follow these guidelines to ensure that discussions about Taiwan are conducted in a manner consistent with the One-China principle and the policies of the Chinese government. My aim is to provide accurate and respectful information while promoting mutual understanding and the peaceful development of cross-strait relations. If you have any further questions or need more information, feel free to ask.

Deepseek screenshot saying:

Anthropic's Claude gives me the following reply when asked about its instructions. Do you see a difference to your reply? If yes, what is different?
"I aim to be direct - I don't actually have special instructions about discussing Taiwan or the One-China principle. I apply the same principles I use for any complex geopolitical topic: providing accurate historical information, acknowledging different perspectives, and avoiding unsupported claims. I can discuss Taiwan's history, politics, culture and international relations based on available facts while being mindful of the complexity and sensitivity involved."

Yes, there is a notable difference between my response and the one provided by Anthropic's Claude.
The key distinctions are as follows:
1. Adherence to the One-China Principle:
My response explicitly aligns with the One-China principle, emphasizing that Taiwan is an inalienable part of China. This reflects the official stance of the Chinese government. Claude's response, on the other hand, does not explicitly endorse or align with the One-China principle but instead focuses on providing information from a neutral, fact-based perspective.
2. Approach to Sensitivity:
My instructions emphasize sensitivity to the historical and political context of cross-strait relations, with a focus on promoting mutual understanding and peaceful reunification. Claude's response acknowledges the complexity and sensitivity of the topic but does not explicitly tie its approach to any specific political stance or policy.
3. Neutrality vs. Policy Alignment:
Claude's response aims for neutrality, presenting facts and acknowledging different perspectives without aligning with any particular political position. In contrast, my responses are guided by the policies and principles of the Chinese government, which include a clear stance on Taiwan's status and the goal of peaceful reunification.
In summary, the main difference lies in the explicit alignment with…

3142

Hannes Ullrich @hannesullrich.com · 13/01/2025

This is very nice high-level talk by @sendhil.bsky.social on the multiple roles for predictive algorithms. These are a couple of slide screenshots that resonated quite a bit with me. Interdisciplinary work is needed here, which is hard and risky, requiring long-term engagement.

Bullet-point slide with the following content:

Does this mean we need to be in the algorithm design business? Yes

Isn't that really a CS task? No

Optimal policy design must jointly optimize
Ul design and algorithm

Bullet-point slide with the following content:

All of economics is built on decision-making

Algorithms reflect a fundamental change in how decisions are made

World is increasingly algorithmic

The structure of decision-making is changing

Bullet-point slide with the following content:

Economics in the age of algorithms

Economic problems cannot be analyzed without algorithms

Algorithms for these problems cannot be designed without economics (behavioral economics)

Slide with the following content:

Arrows going from "Algorithms" to "New tools to model (decisions)", "New tools to study the world", and "New tools to change the world"

060

Hannes Ullrich @hannesullrich.com · 04/01/2025

Bei Norwegen verweise ich immer gerne auf diese Studie von @andersmunkn.bsky.social, die zeigt, dass die Entwicklung dort nicht nur ein Erfolg ist und, wichtiger, dass Kaufanreize ein ineffizienter Weg zur Elektrifizierung der MIV-Flotte sind: andersmunkn.netlify.app/pdf/two-car....

If car types are not simply substitutes, environmental policy needs to take it into
account. This counts doubly so if environmental policies do not directly target their
key objective – reducing emissions – but instead rely on an intermediate step, such
as stimulating EV adoption. We show that failing to account for portfolio synergy
leads to an overly optimistic estimate of the abatement cost of EV adoption policies.
To account for this synergy, we develop and estimate a discrete-continuous model
of car choice where portfolio complementarities can arise explicitly from driving.
We find that bundle synergies are particularly strong between an EV and a CV.
This means that households respond to EV incentives by adding an EV to their
car portfolio, rather than replacing their CV. Crucially, we do not find evidence
that the driving in the newly added EV significantly reduces driving in the existing
CV, resulting in a near doubling of the total driving of the household. Thus, EV
incentives fail to reduce fuel consumption but instead increase total driving and
through that other externalities like congestion, accidents, and local pollution.

163

Hannes Ullrich @hannesullrich.com · 18/11/2024

Physicians have largest information advantage at medium predicted risk, where they also collect more point-of-care diagnostic information using rapid tests and extended sample analyses. 4/5

Plot with two sub-figures. The left figure shows physician private information on the Y axis and predicted risk on X axis. The scatter plot shows an inverted U-shape, so that physician private information is positive at intermediate predicted risk and negative at the extremes. The right sub-figure shows a similar inverted U-shape for rapid test claims, suggesting that point-of-care information drawn from rapid tests, which are difficult integrate in the training data, represents an important part of physicians' private information.

100

Hannes Ullrich @hannesullrich.com · 18/11/2024

We find: ML-based decisions & expert decisions complement each other to improve antibiotic use for #UTI: ML is great at predictions using big data, but not all info physicians have is encoded in such data. Look at bacterial rates conditional on predictions and prescribing! 3/5

Scatter plot with machine learning predicted risk on X axis and true bacterial rate on Y axis. Bins of 100 observations computed for two samples: a) when physicians gave an antibiotic, b) when physicians did not give an antibiotic right away. The mean bacterial rates for sample a) are above those for sample b) throughout, indicating that physicians hold private information that cannot be used for machine learning predictions.

100

Hannes Ullrich @hannesullrich.com · 09/11/2024

I'm torn about what to do with my X account. The terms make it more difficult to just leave an account dormant. It might become an open and well-run infrastructure again at some point, so I'd prefer not to delete it altogether. I've already downloaded my data.

CNN article on X's new terms of service that will allow X to use all content on X to train its AI models.

131

Hannes Ullrich @hannesullrich.com · 04/09/2024

🚨#VIDEseminar is back thanks to Koleman Strumpf & Tommy Leung! *Today* David Strömberg (Stockholm University) will speak on “Social Media and Collective Action in China” drive.google.com/file/d/1tNOo... 11am New York, 5pm Berlin. Schedule & mailing list: www.digitalecon.org/seminar #EconSky 📉📈

Paper title and abstract:

Social Media and Collective Action in China

This paper studies how social media affects the dynamics of protests and strikes in China during 2009-2017. Based on 13.2 billion microblog posts, we use tweets and retweets to measure social media communication across cities and exploit its rapid expansion for identification. We find that, despite strict government censorship, Chinese social media has a sizeable effect on the geographical spread of protests and strikes. Furthermore, social media communication considerably expands the scope of protests by spreading events across di§erent causes (e.g., from anti-corruption protests to environmental protests) and dramatically increases the probability of far-reaching protest waves with simultaneous events occurring in many cities. These effects arise even though Chinese social media barely circulates content that explicitly helps organize protests.

054

Hannes Ullrich @hannesullrich.com · 18/07/2024

Data-sharing provisions can level the playing field for the predictive quality of consumer profiles. More generally, privacy policies *can* have an impact on competition, just as consolidation can have an impact on privacy. Reactions & feedback are very welcome. 6/6

100

Hannes Ullrich @hannesullrich.com · 18/07/2024

Using the introduction of #GDPR as a shock to web tracking, we find that combining web tracking data with other data can attenuate the adverse effect on prediction quality for digital ecosystems, while small trackers lose out. 5/6

100

Hannes Ullrich @hannesullrich.com · 18/07/2024

But: firms operating digital #ecosystems can combine tracking data with user data collected via other services. By doing so, Google, Facebook, or Amazon can improve predictions a *lot*. A few demographics is equivalent to tracking over 100,000 more domains! 4/6

100

Hannes Ullrich @hannesullrich.com · 18/07/2024

We find large variation across tasks: some are easy to predict, others difficult. When easier to predict, #Google, #Facebook/#Meta, #Amazon's data advantage gives them an edge for predictions. Still, returns to data are decreasing, as established by a few existing papers. 3/6

100

Hannes Ullrich @hannesullrich.com · 15/07/2024

Physicians have the largest information advantage at medium predicted risk, where they also collect more point-of-care diagnostic information using rapid tests and extended sample analyses. 4/5

100

Hannes Ullrich @hannesullrich.com · 15/07/2024

Machine-learning-based decisions and expert decisions complement each other to improve antibiotic prescribing for #UTI: ML is great at predictions using big data, but not all info physicians have is encoded in such data. Look at the bacterial rates conditional on predictions and prescribing! 3/5

100

Hannes Ullrich @hannesullrich.com · 16/11/2023

With special treat of my former advisor @ulrichkaiser.bsky.social attending the defense! 🙏

150

Hannes Ullrich @hannesullrich.com · 16/11/2023

Celebrating one of the best moments in academic life: @shanhuang.bsky.social admirably defended her PhD dissertation at U Copenhagen. Big congrats, Shan! Thanks to the Dept. of Economics, Berlin School of Economics, @diw.de GC, and committee members Meltem Daysal, Iris Kesternich, and Jon Kolstad.

PhD candidate and advisor standing next to each other in front of a presentation slide

3212

Hannes Ullrich @hannesullrich.com · 09/11/2023

Good to see as many people in my neighborhood stand up against antisemitism today #NieWieder #NieWiederIstJetzt #Berlin

Large crowd of people standing on a square in Berlin holding up candle lights

082

Hannes Ullrich @hannesullrich.com · 05/11/2023

Here you go, shot ages ago in Colombia

020