iiinova — where your digital self keeps evolving

iiinova is part of the Atmosphere — an open, public network where you own your identity.

Ivo Verhoeven

@ioverho.eurosky.social

59 followers 562 following 5 posts

PhD candidate @ Institute for Logic, Language and Computation, University of Amsterdam www.ivoverhoeven.nl

Posts Replies Media

Ivo Verhoeven @ioverho.eurosky.social · 14/08/2026

A big thank you to the editors, editorial assistants, and reviewers at @complingjournal.bsky.social. This was (by far) the best experience I've had with peer review so far.

000 0

Ivo Verhoeven @ioverho.eurosky.social · 14/08/2026

It’s our hope that misinfo-general encourages more realistic evaluation practices and a rethinking of what constitutes a 'good' misinformation detector in the first place. Actually solving this problem requires more than high scores on academic benchmarks.

100 1

Ivo Verhoeven @ioverho.eurosky.social · 14/08/2026

The misinfo-general benchmark uses a relaxed definition of misinformation, allowing us to capture a long-term snapshot of content produced by unreliable sources. We then show how one can simulate a large set of distribution shifts using content and publisher metadata.

100 1

Ivo Verhoeven @ioverho.eurosky.social · 14/08/2026

How are we supposed to detect tomorrow's misinformation when all we have is yesterday’s news? We want inductive classifiers that can operate in low context settings while being robust to severe distribution shifts, but how do we train and evaluate for these properties without adequate data?

100 1

Ivo Verhoeven @ioverho.eurosky.social · 14/08/2026

Besides being The Problem of our times, detecting misinformation is also a fascinating engineering challenge. It's non-stationary, has no clear definition, and labels are mostly context-dependent. Misinformation *datasets* are none of these things. This article is dedicated to that dichotomy.

120 2