RE: @rwthaachenuniversity@wisskomm.social
another German university joins Mastodon. Say hello!
Remote
Ulrike Hahn
@UlrikeHahn@fediscience.org
mastodon 4.6.9Academic @Birkbeck, Univ. of London
Centre for Cognition, Computation, and Modelling
was just at Netherlands Institute for Advanced Study working on Digital Democracy with Davide Grossi and Michael Maes
and then at the Munich Centre for Mathematical Philosophy, LMU
now back in London
works on #rationality #argumentation #testimony #SocialNetworks #misinformation #ComputationalSocialScience #DigitalDemocracy
2484 Followers
1863 Following
50 Posts
Joined November 07, 2022
Research Centre:
cccm.bbk.ac.uk
SciBeh Initiative:
SciBeh.org
Since taking office in January 2024, the current US administration has terminated over 100 scientific advisory committees
https://www.nature.com/articles/d41586-026-01301-5
a revealing experiment in AI based data analysis……
https://kucharski.substack.com/p/real-signals-or-artificial-stereotypes
I recommend (for multiple reasons) this piece on coding with AI (Claude Code):
the end of online anonymity?
I see virtually no discussion of the relationship between protocol design and carbon footprint…
shouldn’t that be one of the most important considerations in thinking about social media design right now?
good piece!
“the development and application of methods, the training of minds, the creation of people who know how to think about hard problems. If you hand that process to a machine, you haven't accelerated science. You've removed the only part of it that anyone actually needed.”
Replying to
the final presentation, via Zoom (the only one not in person…), was by Imran Ahmed the founder and CEO of the Center for Countering Digital Hate who can now no longer leave the US where he lives because he would not be allowed re-entry as one of the individuals working on misinformation and digital hate that received visa sanctions by the US State Department in Dec 2025 ….and that in itself was a poignant reminder of what is currently at stake in trying to build online tools and systems that can genuinely serve democratic societies
https://www.bbc.co.uk/news/articles/cp39kngz008o
on the shifting nature of social media: social media are now para social media
interesting paper on how training models on (partial) output from other models transfers unexpected (and difficult to track) additional representational content: this can amplify unwanted signals/behaviours even through training data that seem wholly unrelated to that preference/behaviour
this occurs due to the interaction between gradient descent as a training procedure, shared initialisation, and the fact that internal representations involve super-position #LLM
Request for advice #accessibility
I am looking for a tablet based solution that ideally satisfies the following constraints for a hard of hearing person:
1. high quality speech transcription
2. works for *groups of speakers*
3. ideally different speakers are colour coded
4. simple to use & easy to read full screen display
5. also works offline
I can satisfy some of these with tools I know of, but not all -most readily droppable features are 5 and 3.
any suggestions?
this is a beautiful exposition of what science is, both in content and writing
https://lisagraumlich.substack.com/p/the-unfinished-sentence
(via Liz Neeley https://www.liminalcreations.com/newsletter)
Replying to
@douginamug … it feels to me like where we’re headed is that all of this is becoming so easy anyone is at risk, and that risk is determined simply by the probability that someone holds a sufficient grudge…
Replying to
@hnnng new Mastodon server with 16 large German funders launched....
https://www.bosch-stiftung.de/en/press/2026/05/16-german-foundations-take-stand-digital-sovereignty-and-freedom-expression
Replying to
@mxp@mastodon.acm.org @tomstoneham@dair-community.social interesting paper….am a bit perplexed by the reversability constraint as implemented: instructions such as “add Rembrandt style lighting” and “remove Rembrandt style lighting” seem too vague to me to guarantee uniqueness?
Replying to
@stefanmuelller is this the work in question? https://aclanthology.org/2024.acl-long.800.pdf
science….
Replying to
@ngaylinn@tech.lgbt it’s a great example, isn’t it… will use it for this week’s lecture, with the supplementary discussion question: “so what, if anything, makes you confident that it is not expressing the same kinds of bias in your requested summaries and explanations of research articles?”.
Replying to
@skribe I’m sure it was just a year ago (though it feels longer), are you saying we are already in 2026 now?
Replying to
@mxp@mastodon.acm.org @tomstoneham@dair-community.social I guess my intuition is that unless the reversal is a strict “undo” nothing about the results is a) surprising and b) limited to LLMs nor even c) meaningfully classed as an error…. so it’s really all down to the actual operations themselves, so it’s all in the detail (and if “add Rembrandt lighting” really is an example they used, I see problems)
Replying to
@mxp@mastodon.acm.org @tomstoneham@dair-community.social …for what it’s worth, I also thought “oh this is a clever operational test” but then I looked at the methods section and now am really not sure
Replying to
@mxp@mastodon.acm.org @tomstoneham@dair-community.social actually, I think I have now understood the source of (my!) error - the “chain” consists of strung together immediate forward/backward steps not concatenation of steps and their undoing?
so linear transformations would be fine as one is only ever doing and undoing one operation at a time?
that would still leave the original problem of potential ambiguity in the instructions (add/remove Rembrandt lighting?)
(though non-commutativity across individual operations would still play an indirect role as it would amplify the emergence of differences, ie it makes it the case that errors interact?)
Replying to
@tomstoneham@dair-community.social @mxp@mastodon.acm.org hmmm… feels more like a potential problem with the method of the paper to me?
the conceptual core is the idea that all operations are reversible and symmetrical so do not give rise to order effects, but I don’t see how that could be the case?
unless the reverse operation is strictly “undo the specific Rembrandt lighting you added” removing Rembrandt lighting need not take you back to the same place, and there can then also legitimately be order effects….
and *if* the reverse operation is “undo the specific Rembrandt lighting you added” those “undos” would necessarily *only* work in reverse sequential order (by definition)
either way I don’t understand how the actual test lives up to the expressed constraints!
or am I missing something?
Replying to
@tomstoneham@dair-community.social @mxp@mastodon.acm.org I suspect 2) is simply impossible-
even very basic, reversible, transformations (such as scaling, rotation, reflection) aren’t commutative!
(am still thinking this is so obvious, I must have misunderstood something…)
Replying to
@laurenshof imagine if instead of the current protocol wars and arguing about the meaning of the word ‘decentralisation’, we actually raced and competed to make our tools more energy efficient …..
the vibe coded social network seems to be down again…..
really not selling the approach
Replying to
@pbloem @ngaylinn finally, I suspect the confusion is fuelled by the fact that there is a form of inference (deduction) that is 'truth preserving' but not only does that still require that premises are true, it's also very obviously the case that the overwhelming majority of inferences we conduct as humans are not deductive
Replying to
Replying to
@sigismundninja “Our contributions. We demonstrate that LLMs funda-
mentally change the picture, enabling fully automated deanonymization attacks that operate on unstructured text
at scale. We show this by phrasing deanonymization as a matching problem and showing LLMs can perform all steps needed to match accounts: extract identity-relevant signals from arbitrary text, efficiently search over millions of candidate profiles, and reason about whether two accounts belong to the same person. We show that the practical obscurity that has long protected pseudonymous users (the assumption that deanonymization, while theoretically possible, is too costly to execute broadly) no longer holds”
Replying to
@Steve_Lindsay Steve, won’t the way you route, aggregate and store content translate directly into energy used?
Replying to
@wwhitlow I think the bigger problem is that a preprint server isn’t itself meant to provide peer review- it’s not a journal and the process isn’t designed for that.
Replying to
@mackuba@martianbase.net yes, I understand the differences, but I do think the “they do not exist” answer is becoming increasingly wrong, particularly if these stack develop their own stack specific governance, like Blacksky has done.
Replying to
@Enthalpiste @wwhitlow well that was the idea behind publishing reform: to move from the bottleneck of journals to a kind of crowd sourced evaluation. But that shift is exactly what is breaking here. arXiv isn’t a journal, it’s a preprint server, that is, the first step in the new architecture that was meant to make those transitions away from traditional journals possible…
the underlying problem is scale… there are only so many human experts available to the system… and we’ve moved beyond their capacity
Replying to
@npub147fp2j606qpfysp38phhzvempt7ewsdqwm6uww9uycp6tdvavu0s5paq44
one way we have of thinking about this might be distinguishing between the things we feel, as scientists, we all needed to learn (e.g. design an experiment, programme an experiment, do the statistical analysis etc.) but don’t feel we need to do ourselves all the time in later projects and the level of involvement that is always required meaningful contributor to the project?
“Bob” is a problem because he is never even meaningfully learning to do things. Likewise, the “senior” contributor who never gets stuck into any of the detail is probably also a bit of a problem
Replying to
@mxp@mastodon.acm.org @tomstoneham@dair-community.social interesting that it would seem intuitive to a computer scientist given that function composition is typically non-commutative…
Replying to
@pbloem @ngaylinn or to put it differently “fact checking” is not an agent internal process in the first place and Tufecki’s quote seems to indicate confusion between agent internal processes for coming up with a putative answer with the raft of external tools and actions we have for checking or maintaining ‘truth’
Replying to
@MsMerope I think it runs a bit deeper than that and isn’t just driven by him….the attempt to silence and/or control science is a feature of authoritarianism more generally
https://blog.ucs.org/melissa-finucane/science-caught-in-a-rip-tide-how-authoritarianism-sweeps-away-evidence-based-policy/
Replying to
Replying to
@jon_bon on my understanding, ATproto is a protocol, that is, a bunch of standards and rules (for example what the basic data structures look like). That standard is conceptually distinct from specific software that makes use of those specifications or standards - for example Bluesky’s appview
It’s Bluesky’s software that is apparently increasingly developed with Claude, but that isn’t something inherent in ATproto …it’s just something the developers at Bluesky PBC the company choose to do (along with all the concerns such as energy consumption that raises).
https://en.wikipedia.org/wiki/Communication_protocol
Replying to
@ngaylinn @pbloem and this is one of the earliest (?) papers on calibration in LLMs
at the same time, calibration in humans has been the topic of many decades of study within judgment and decision-making research….
https://arxiv.org/pdf/2207.05221
Replying to
@mxp@mastodon.acm.org @tomstoneham@dair-community.social Michael, that I think would be more problematic (and what I originally assumed) whereas what I now *think* (?) they are doing is Markdown -> docx -> Markdown; Resultant Markdown -> Latex -> Markdown; Resultant Markdown -> troff -> Markdown
(the comment about order independence in the set of reversible operations is still weird though…)
Replying to
@mxp@mastodon.acm.org @tomstoneham@dair-community.social “I don't know much about image editing, so I assume you are right.”
no need to believe me, Tom- here’s a simple example: imagine an irregular object of your choice, now imagine stretching it along the horizontal axis, then rotate it. Contrast the result with the order 1) rotate 2) stretch….
Replying to
@mackuba@martianbase.net there *were* no instances in the original conception but aren’t the developing independent complete stacks Bluesky, Gander, Eurosky etc. effectively recreating some sort of instance?