Wayback arxiv is your friend against impression management

news
Author

Dimitra Maoutsa

Published

September 23, 2026

In the past days I became aware of this PR announcement from Tuebingen AI account regarding an additional new member in their directorship.

Notice the time of the post: 6:11 PM 8 July 2026.

Some hours earlier, I had posted on bluesky this post, announcing the presentation of my posters at the ICML conference

Notice the time of the post: 5:22 AM 8 July 2026.

Incidentally, the paper I was presenting in the main conference (which was the last paper from my PhD thesis that has been rejected for a year from all the big conferences), was relying heavily on a framework developed by Georgios Arvanitidis in the lab of this newly announced director in Tuebingen AI center. And I was and am as grateful as I couldn’t ever be, that this framework existed when I needed it (and I happened to stumble upon it), because otherwise I would have to develop it from scratch or use some other approaches that didn’t work so well for constructing geodesics on empirical manifolds.

At some point when I was writting this last chapter of my thesis and the associated workshop paper, I though to reach out to the authors to ask whether they would be willing to collaborate on this with me or be co-authors on my paper (given that I have been using their code). However given their connection to Tuebingen, I didn’t, and finished the framework alone, and only cited the work and thanked Georgios for keeping a well maintained repository that was easy to integrate with the rest of my work.

Now this might seem just like a random coincidence, that at the day when I was presenting my relevant work, Tuebingen AI announced the new director, who happened to be the senior author of a paper that was instrumental for my last PhD project.

Of course since Tuebingen AI and other accounts connected to Matthias Bethge are blocked from my bluesky account (see here https://dimitra-maoutsa.github.io/values/) due to the problems he keeps creating for my career for years now (for reasons he only knows…), I had absolute ingorance up until the last days about this announcement.

However, what I knew already from a year ago, was that this new director was appointed as director to the Tuebingen AI center one year earlier, on July 2025, exactly at that period when this (now) ICML paper was under review for Neurips from which it got rejected.

And wayback arxiv is our friend here because you can easily go back to the “Peoples” page from Tuebingen AI during 2025 and clearly see when that person became a director. First version below is from 1st of June 2025 (notice the date on upper right corner)https://web.archive.org/web/20250601094617/https://tuebingen.ai/people

and second is from 1st of August 2025 (notice again the date) https://web.archive.org/web/20250801105350/https://tuebingen.ai/people

Why the PR of this director appointment happen to get announced one year later, at the exact same date when my poster got presented, is something that Matthias Bethge and Tuebingen Ai know.

From my side, to actually get my paper accepted I had to ensure that no person who might have been under the influence of Matthias Bethge would have any decision making role on my submission, otherwise the paper would continue getting rejected.

And to clarify here, I have absolutely no issue with any of the people appearing here (excluding Matthias) in the screenshots, and I couldn’t be more grateful for the existence of the work of Georgios Arvanitidis and Philipp Hennig at the time when this was instrumental for finishing my PhD.

Why Matthias is engaging in this sort of impression management, is something only he and my former postdoctoral advisor know! You can read more here at the bottom about how old their relationship (values)


Also while my paper was under review, some person who appears to be an undergraduate(!) has been submitting this paper https://openreview.net/forum?id=aVGg6lRCVi&referrer=%5Bthe%20profile%20of%20Gao%20Jiang%5D(%2Fprofile%3Fid%3D~Gao_Jiang1) at least in ICLR and ICML (where I happened to review it), which contained unreasonably exotic terminology, which nevertheless was using terms that were making it similar to my submission, and was broadly postulating the reverse approach from my paper, although it was really refering to state dependent deiffusions, without explicitly stating it. The paper was almost about to be accepted in ICLR (!), which makes absolutely no sense, and fortunatelly I happend to review it at ICML and called out the author about the incoherent writting and bogus contribution, to which they really didn’t have a lot to reply, and stepped back. I still do not beleive that an undergratute student was behind the paper, because although the contribution was non-existent, the terminology and the mathematical background require some courses in stochastic analysis that I doubt an undergraduate would have had the time to take and then write a paper on top of that.

This was part of my final assesment of that paper in ICML, and I still think the second paragraph might be true:

Overall after reading the paper, and the rebuttals, I feel that novelty is overstated, the terminology was artificially exotic, and the experiments unclear what they were trying to demonstrate. Thus I propose to reject the paper, and even flag it as a potential product of LLMs.

I even wonder whether the authors deliberately used this exotic/altered terminology to make the submission more similar to papers that had previously been rejected at other venues, in the hope of increasing the chances that both this submission and the targeted papers would be assigned to the same reviewers, or that the authors would have a higher chance of reviewing the targeted papers. Still not sure what to conclude…

One of their arguments, that (state dependent) diffusion induces an additional geometry in the space, is true, and known fact (and somewhat obvious for everyone with experience on dynamical systems), but they do not explicitly point out that they do refer to state dependent diffusion terms. Sometimes in their equations they use \(\sigma\) do indicate the diffusion, which would imply that it is a constant, while their argument holds when the diffusion is a vector field, and thus should be denoted with \(\sigma(x)\).


I am equally eager to see who is behind this paper https://openreview.net/forum?id=HgbFRA7qA5 that seems to go towards acceptance but does not clearly specify what do they connect with the bridges :) In this setting the observations are continuous so not sure what do they mean with ““Given pairs of observation-inferred start and end latent distributions, can we learn to infer the internal dynamics \(f_\theta\) and trajectory-specific control \(v_\phi\) required to explain deviations from these dynamics?” (also they consider Gaussinaa observation process and not poisson but still claim to outperform FINDR in the experiments which uses poisson observation process…) I am just waiting the rebuttal, unfortunately I am not a reviewer.