Lightnews — Scholar-powered news

Sébastien Destercke

@sebastiendestercke.bsky.social

510 followers 180 following 77 posts

CS researcher in uncertainty reasoning (whenever it appears: risk analysis, AI, philosophy, ...), mostly mixing sets and probabilities. Posts mostly on this topic (french and english), and a bit about others. Personal account and opinions.

Posts Media Videos Starter Packs

Pinned

Sébastien Destercke @sebastiendestercke.bsky.social · Nov 26

To share my love (this word seems not too much, as I worked in the field for nearly 20 years now 😅) of uncertainty reasoning with imprecision, I will start a post where I will share some papers I particularly like at a non-regular frequency 😵‍💫. Let us see how this pans out.

1 7

Sébastien Destercke @sebastiendestercke.bsky.social · 2d

The talk of Michele Caprio on connecting conformal prediction, credal sets (a.k.a. convex sets of probabilities) and Bayesian approaches through Category theory is now online at www.youtube.com/watch?v=hLdC....

Seminar-M.Caprio: Conformal prediction and its uncertainty quantification capability via credal sets

YouTube video by Imprecise Probabilities Channel of SIPTA

www.youtube.com

Sébastien Destercke @sebastiendestercke.bsky.social · 3d

French voters/parliament as scientific reviewers/editors : we ask a major revision of your manuscript.

French government as authors : here is a new version. We only corrected the typos.

Reposted by Sébastien Destercke

David Cayla @davidcayla.bsky.social · 5d

1/3
La fonction publique meurt du fait de la perte de sens des métiers, d'une bureaucratisation excessive et d'un niveau de rémunérations inférieur à celui du marché.

Pendant ce temps, le débat se focalise à gauche et à droite sur la fiscalité et les allègements d'impôts.

3 16 30

Reposted by Sébastien Destercke

Sébastien Tixier @sebtixier.bsky.social · 5d

CVAE : « Le gouvernement entend baisser un impôt de production qui pèse principalement sur les PME » selon l'entourage de Lecornu.
La CVAE concerne principalement les PME ??
Ces gens nous prennent vraiment pour des cons.
Et quand il y a entourloupe...⤵️
www.leparisien.fr/politique/bu...

Budget : Sébastien Lecornu confirme une baisse d’un impôt de production, la CVAE, pour 1,1 milliard

La CVAE, un impôt de production pesant sur les entreprises, était censée disparaître intégralement en 2024, mais sa suppression a été report

www.leparisien.fr

3 35 41

Sébastien Destercke @sebastiendestercke.bsky.social · 6d

Mon impression: souvent les services pensent bien faire et eventuellement nous aider, mais n'ont aucune idée du nbre total de sollicitations de ce type que nous avons.

Sébastien Destercke @sebastiendestercke.bsky.social · 6d

Vis ma vie de chercheur: fête de la science, formulaire 10 pages (2x); CSI 4 pages + lettre recommandation ; mission: création d'agents, feuille de mission, logiciels; sujet/bourse de these: presque même info sur 3 formulaires différents pour chaque instance. etc

Reposted by Sébastien Destercke

CNRS @cnrs.fr · 12d

Global warming is not a “con job.” It is a fact.

In the face of Donald Trump’s new denial of climate change, CNRS scientists have reacted by pointing to the facts and levers for action to halt this planetary crisis.

Global warming is not a “con job.” It is a fact.

In the face of Donald Trump’s new denial of climate change, CNRS scientists have reacted by pointing to the facts and levers for action to

www.cnrs.fr

2 23 36

Sébastien Destercke @sebastiendestercke.bsky.social · 16d

Les arguments volent dans tous les sens quand on est dans l'oreille du cyclone!

Sébastien Destercke @sebastiendestercke.bsky.social · 16d

Interested in how conformal prediction, machine learning and uncertainty quantification mix together? Then do not miss our first SIPTA virtual seminar of the year this 1st October (3pm CEST), given by Michele Caprio. More info at sipta.org/events/sipta....

SIPTA Seminars | The Society for Imprecise Probabilities: Theories and Applications

sipta.org

2 5

Reposted by Sébastien Destercke

Feena van @feenavan2099.bsky.social · 16d

"Italian labor unions led a massive 24-hour general strike on Monday to protest Israel’s ongoing genocide in Gaza, with estimates of hundreds of thousands of demonstrators rallying in dozens of cities across Italy."

Italy's Unions Lead General Strike for Gaza | Common Dreams

"Meloni should take a stand with the facts against those who have slaughtered 20,000 children, rather than limiting herself to saying 'I do not agree,'" said one critic of Italy's right-wing prime min...

www.commondreams.org

3 90 310

Sébastien Destercke @sebastiendestercke.bsky.social · 17d

😅 Dans ces cas là je mets dans mon retour aux auteurs qu'ils doivent se sentir libre de faire un choix (argumenté) parmi les citations demandées. J'ai souvenirs d'une publi mentionnant explicitement les références inutiles demandées par les réflecteurs dans une section.

Sébastien Destercke @sebastiendestercke.bsky.social · 17d

I still have nightmares from a Paris Vienna with three changes in Germany 😱

2 1

Reposted by Sébastien Destercke

José Cerca @naturalselection.bsky.social · 19d

7 55 250

Sébastien Destercke @sebastiendestercke.bsky.social · 19d

The positive answer is that giving constructive feedbacks to the authors is useful if they care about it, and that reviewing should be primarily about that? the less positive one is that with 21000 papers, final decision beyond mere thresholding becomes difficult, if not impossible.

Sébastien Destercke @sebastiendestercke.bsky.social · 23d

TIL that in an older version, Cinderella had a first step-mother that she killed upon the suggestion of the second one (that had 2 daughters), and received her dress from a magic tree 🎄.

Learned it from "Et à la fin, ils meurent: La Sale Vérité sur les contes de fées" by Lou Lubie (in French)

Sébastien Destercke @sebastiendestercke.bsky.social · 24d

Not sure there is many room for debate, sorry 😅. The tones are different but I would say our conclusions are pretty similar? The fact is that those AI-generated reviews are much better than I initially anticipated. Despite their obvious defaults, I remain impressed by the result.

Sébastien Destercke @sebastiendestercke.bsky.social · 24d

but not to an expert that can help you improve the core of the work and be a trustable critic. Hence my opinion that it may be helpful to the authors if they could have it before submission.

1 2

Sébastien Destercke @sebastiendestercke.bsky.social · 24d

I do think that they can be helpful to polish the paper in some parts and possibly to identify pieces of literature that may be relevant (after a thorough check). In summary, I associate them to a quite reasonable proofreader with access to a wide range of literature,

1 1

Sébastien Destercke @sebastiendestercke.bsky.social · 24d

Hello @mcognetta.bsky.social, I "mostly" 🙃 agree with what you say, especially that they are currently useless for decision making, and are not very helpful to improve the actual research, both things requiring the perspective/nuance I mentioned.

1 2

Reposted by Sébastien Destercke

Gemma Clark 🇵🇸🇵🇸🇵🇸 @gemmaclark.bsky.social · 26d

Who killed Charlie Kirk? A white man, raised Christian with Republican parents. Tyler Robinson donated to Trump and later became a ‘groyper’, ie a neo-Nazi. Predictable. And guns are still the problem www.thecanary.co/trending/202...

Charlie Kirk suspect is a gun-loving Trump supporter

From what we know so far, the suspect seems to be part of the 'groyper' movement of far-right online agitators

www.thecanary.co

3 38 87

Sébastien Destercke @sebastiendestercke.bsky.social · 29d

Especially as I was rather skeptical at first of these AI-generated reviews :).

Sébastien Destercke @sebastiendestercke.bsky.social · 29d

As SPC I did not use them to make recommendations. As an author I would have checked them for improvement, and as a reviewer I would have found them to be a good support.

I did not say they were/were not the solution, just that I was rather amazed at the result, and that it raises many questions.

1 1

Sébastien Destercke @sebastiendestercke.bsky.social · 29d

Beyond the fact that big A* conferences reviewing is not reliable, I do not know what to think either. This does not prevent good contributions to be there and the field (AI) to advance, just that big conferences are becoming like in other fields (we just do not really acknowledge that)?

Sébastien Destercke @sebastiendestercke.bsky.social · 29d

The press release aaai.org/aaai-launche... does not really say anything about it, but there will be a detailed analysis. Examples on publicly available papers (e.g., on arXiv) would probably be better as submitted papers should remain private? Not sure how open the system will become.

AAAI Launches AI-Powered Peer Review Assessment System - AAAI

AAAI, a leading nonprofit dedicated to advancing scientific research and collaboration, launches AI-Powered peer review assessment system.

aaai.org

1 1

Sébastien Destercke @sebastiendestercke.bsky.social · 29d

It also confirms my impression that big AI conferences are just becoming big showrooms, probably useful to exchange and showcase some selected works, but rather inefficient in terms of scientific quality screening and useful off-event feedbacks. 4/4