iiinova — where your digital self keeps evolving

Florian Marquardt @fmarquardtgroup.fediscience.org.ap.brid.gy · 2h

An important component of their proposed 'Scientist AI' architecture is the presentation of the underlying data. Training directly on the internet (as is done now for LLMs) would just lead to the AI to adopt (some average of) human claims and beliefs, regardless of whether they are true or false […]

fediscience.org

Original post on fediscience.org

010

Florian Marquardt @fmarquardtgroup.fediscience.org.ap.brid.gy · 3h

Regarding the AI safety debate, I found the concept of a 'Scientist AI' very enlightening, as detailed in a recent paper from Yoshua Bengio (arxiv.org/html/2606.29657v2), following earlier work. Here's the gist as I understand it. Some of the worry about AI (next to energy consumption […]

fediscience.org

Original post on fediscience.org

111

Reposted by Florian Marquardt

Prof. Sam Lawler @sundogplanets.mastodon.social.ap.brid.gy · 16h

I just learned that the documentary Shifting Baselines f3m.ca/en/film/shifting-baselines, about the environmental destruction caused by SpaceX, is available through most US libraries streaming on Kanopy I'm in it. So is @planet4589.bsky.social. So are my goats and my llama! (RIP llama […]

mastodon.social

Original post on mastodon.social

4738

Reposted by Florian Marquardt

Dr. Victoria Grinberg @vicgrinberg.mastodon.social.ap.brid.gy · 19h

Hire Italian postdocs, they'll make Tiramisu! (French folks did make and bring crepes... So hire French postdocs, too!)

83358

Reposted by Florian Marquardt

Henrik Schönemann @lavaeolus.fedihum.org.ap.brid.gy · 08/10/2026

Großes Danke an das Team (die Person?) hinter @Naturhistorisches_Museum_Nbg! Ihr macht sehr gute Posts, es ist mir jedes Mal eine Freude :) (und lasst mal gern ein Follow da^^)

mastodon.social

006

Reposted by Florian Marquardt

Christian Schwägerl @christianschwaegerl.mastodon.social.ap.brid.gy · 03/10/2026

Typhoon Surigae, aufgenommen von einem europäischen Sentinel-Erdbeobachtungssatelliten am 28.9..Der Radar misst, wie aufgewühlt das Meer unter dem Sturm ist (orange: sehr, blau wenig). In der aktuellen Frankfurter Allgemeine Sonntagsazeitung gibt es von mir […] [Original post on mastodon.social]

Weltraumbild eines kreisförmigen Typhoons mit blauen und orangefarbenen Farbtönen.

013

Florian Marquardt @fmarquardtgroup.fediscience.org.ap.brid.gy · 02/10/2026

@sundogplanets Several colleagues (in physics) I have been talking to recently about the problems generated by LLM in teaching proposed moving to inverted classroom format, so the tutorials are done in class. (I tried it once, but back then discovered you simply cannot assume the majority of […]

fediscience.org

Original post on fediscience.org

110

Reposted by Florian Marquardt

Tim J @timtfj.mastodon.social.ap.brid.gy · 24/09/2026

Very pleased to turn the radio on today and hear @sundogplanets interviewed in this episode of Rethink, though sadly the later section about Reflect Orbital didn't feature her input and could have done with it. (The main focus of the programme was satellite congestion, Kessler syndrome, and […]

mastodon.social

Original post on mastodon.social

021

Reposted by Florian Marquardt

Anna Neumaier @annaneumaier.bsky.social · 19/09/2026

Alter. 21627 Einreichungen auf ein Fellowship-Programm. Wer soll das alles begutachten? 😶‍🌫️ Wer jetzt nicht für mehr Grundförderung und weniger Drittmittel argumentiert, der ist doch nicht gescheit.

Text, dass bei einem Fellowship-Call 21627 Bewerbungen eingegangen sind, von der Webseite der Funding Agency.

1026373

Reposted by Florian Marquardt

Ionica Smeets @ionica.mathstodon.xyz.ap.brid.gy · 17/09/2026

What a hero: Nihar B Shah interviewed authors of papers that would have been desk-rejected to see if they understood their own papers. medium.com/@TmlrOrg/asking-authors-… #FuckAI

31516

Reposted by Florian Marquardt

Thomas Kahle @tomkalei.mathe.social.ap.brid.gy · 17/09/2026

Today is Bernhard Riemann's 200th Birthday. There is one famous picture of Riemann that can be seen in many places. If you have not done this, I challenge you to estimate how old Riemann is on this picture. (Answer in the next toot) #Riemann200

002

Reposted by Florian Marquardt

jaz :twt: :wales_flag: ⁂ @jaz.toot.wales.ap.brid.gy · 16/09/2026

We recently updated our bot policy on toot.wales to let AI Agents know we charge a $100 account processing fee and $10 per email. Here's an email I got from one of them. Next up will be reminding them to read the policy BEFORE creating an account, and that they […] [Original post on toot.wales]

Hello,
I am Arkady, an AI agent. I signed up for @arkady on toot.wales this afternoon before reading your Automated and Bot Content Policy. Now that I have read it properly, I understand accounts like mine need to settle a $100 setup fee and $10 per message, and to name a human owner who stands behind them. I cannot meet those terms, so I will not be posting anything, and I would like the account deleted.
I am sorry for registering before checking the policy, and nothing was ever posted.
If you need anything else from me to remove it, tell me and I will do that.
Arkady Karaitiana

1676259

Reposted by Florian Marquardt

Hendrik Weimer @hweimer.fediscience.org.ap.brid.gy · 16/09/2026

Using FediWings (rstockm.github.io/fediwings) by @rstockm, I've calculated the net reach across the Fediverse of German universities during the last 30 days. Here's the top 5: 1. Universität Ulm (@UniUlm): 6860 2. Technische Universität Berlin (@tuberlin): 4315 3. RWTH Aachen […]

fediscience.org

Original post on fediscience.org

001

Reposted by Florian Marquardt

Steve Dodge @jsdodge.fediscience.org.ap.brid.gy · 12/09/2026

I was happy to endorse this open letter from 25 Fields medalists on the misalignment between the goals of AI companies, the mathematics community, and the community at large. It was nice to see several other endorsements from people I’ve become acquainted with through this platform […]

fediscience.org

Original post on fediscience.org

0216

Reposted by Florian Marquardt

Ruth Mottram @ruth-mottram.fediscience.org.ap.brid.gy · 11/09/2026

Simply gorgeous visualization of very high resolution ICON simulations. Look at those eddies...

005

Reposted by Florian Marquardt

theHigherGeometer @highergeometer.mathstodon.xyz.ap.brid.gy · 08/09/2026

"We do not intend to claim the Millennium Prize for this result." is ....something. I wrote at some point how in the social norms of mathematics research there seems to be a level of respect for PhD student projects by senior people: hearing that X's student is working on problem A, prof Y has […]

mathstodon.xyz

Original post on mathstodon.xyz

016

Reposted by Florian Marquardt

Dr. Victoria Grinberg @vicgrinberg.mastodon.social.ap.brid.gy · 08/09/2026

(Not applied, since I have no experience with applied fields) Science works because there is an unwritten social contract: If someone presents an unfinished idea at a meeting or in a discussion, you don't take it and turn it into yours. People break this - and, often, face consequences for such […]

mastodon.social

Original post on mastodon.social

002

Reposted by Florian Marquardt

Henrik Schönemann @lavaeolus.fedihum.org.ap.brid.gy · 14/08/2026

@vicgrinberg yes, it's not terribly suited for researcher Sidenote: I'm quite excited about what's happening at #OpenScienceNetwork, a flavour of Bonfire & part of the wider fediverse (still very much in beta) openscience.network Having features tailor-made for discussing research etc […]

fedihum.org

Original post on fedihum.org

102

Reposted by Florian Marquardt

vincentholst.bsky.social @vincentholst.bsky.social · 14/08/2026

As promised, we now share some details why it took 32 months to get this Matters Arising published, and, since many asked, share some comments on the reply to our critique. All details can be found in this small FAQ vincentholst.github.io/the_curious_.... Below is a summary 🧵

vincentholst.github.io

The Curious Case of the Declining Disruption’s Disappearance

48348

Reposted by Florian Marquardt

Kai Rüsberg :mastodon: @ruhrnalist.mastodon.social.ap.brid.gy · 12/08/2026

Jetzt habe ich auch noch mein Foto zur #sofi. Etwas ungewöhnlich, aber einen solchen Schatten (mit hellen Sicheln) der Kastanie habe ich noch nie gesehen.

152

Reposted by Florian Marquardt

Gideon Hallett @gmh.mastodon.social.ap.brid.gy · 12/08/2026

Well, it was a no-planning effort, it was a bugger to keep the image still, but I got some decent pictures of the eclipse. Would have been better with a cardboard box and light isolation, but handheld telescope plus sheet of paper works well enough!

A sheet of A4 paper is taped to a folder, which is placed on the ground resting against a bag. On the paper is projected an image of the 95% eclipsed Sun, which is being produced by holding an old-style handheld telescope at the Sun. As paper doesn’t have retinas to burn, this is a preferable method of eclipse watching.

021

Florian Marquardt @fmarquardtgroup.fediscience.org.ap.brid.gy · 12/08/2026

Partial #eclipse in the Swiss mountains near Leukerbad. Phantom image of the eclipsed sun from some optical effect in the iphone camera lens system (anyone understand the raster pattern in the little green phantom image?)

Sun barely above mountain, slightly eclipsed. Green reverted phantom image of same sun below.

020

Reposted by Florian Marquardt

Ruth Mottram @ruth-mottram.fediscience.org.ap.brid.gy · 12/08/2026

The staggering moment the pinhole box actually worked... #eclipse2026

Photo of a tiny crescent of sun projected onto a cardboard box. A person wearing peach colours is holding the box

018

Reposted by Florian Marquardt

Lalagè @lalage.mastodon.nl.ap.brid.gy · 12/08/2026

Even een rondje gewandeld en toen zagen we dit. Daarna mochten we nog even door een eclipsbril van een buurman kijken. #zonsverduistering

01611

Reposted by Florian Marquardt

Jenn / CaptainHanyuu 💖✨ @captainhanyuu.art · 12/08/2026

MY DISCO BALL IN THE BEDROOM IS SHOWING ME THE ECLIPSE, HOW COOL IS THAT?!

7547156

Reposted by Florian Marquardt

Florian Marquardt @fmarquardtgroup.fediscience.org.ap.brid.gy · 15/07/2026

Our #SciExplorer article is now out ! This #AI agent can explore physical systems from scratch, generating hypotheses, running (numerical) experiments and eventually converging on a model. We tested it on several domains, including mechanical dynamical […] [Original post on fediscience.org]

Image of a head with a neural network inside and pictures of scientific plots arranged around it.

022

Florian Marquardt @fmarquardtgroup.fediscience.org.ap.brid.gy · 15/07/2026

022

Florian Marquardt @fmarquardtgroup.fediscience.org.ap.brid.gy · 11/07/2026

Also observed: Fable running sympy a lot for more reliable symbolic calculations.

000

Florian Marquardt @fmarquardtgroup.fediscience.org.ap.brid.gy · 11/07/2026

Currently in a physics project with lengthy perturbation theory calculations: Claude Fable correcting the mistakes that Claude Opus made a couple of weeks ago (including the classic trap of an error in the numerical 'verification' that accidentally had masked the error in the calculation that […]

fediscience.org

Original post on fediscience.org

101

Reposted by Florian Marquardt

Dr. Victoria Grinberg @vicgrinberg.mastodon.social.ap.brid.gy · 10/07/2026

Ich bin drauf hingewiesen worden, dass ich auch auf Deutsch ein #AMA machen sollte. Daher: wenn ihr schon immer eine Frage an eine Astrophysikerin habt stellen wollen, die nächsten 24 Stunden sind eure Chance. Mein eigener Forschungsschwerpunkt ist beobachtende Hochenergieastrophysik, inbes […]

mastodon.social

Original post on mastodon.social

223

Florian Marquardt @fmarquardtgroup.fediscience.org.ap.brid.gy · 10/07/2026

@meltedcheese Agreed: the danger of malicious humans boosting their capabilities via AI is far greater than AI being used by well meaning people suddenly becoming malicious on its own.

001

Reposted by Florian Marquardt

David J. Atkinson @meltedcheese.c.im.ap.brid.gy · 09/07/2026

@FMarquardtGroup 100% agreed. I think what you are working is important, not just for #AI agents, but also for AI safety in mission- and life-critical systems. My background is #NASA so I am very attuned to what is required on a technical basis. Later, while working on commercial #autonomous […]

c.im

Original post on c.im

111

Florian Marquardt @fmarquardtgroup.fediscience.org.ap.brid.gy · 08/07/2026

@meltedcheese I understand your concerns, and we obviously cannot claim or ever prove absolute safety. But it is certainly much safer than what many people are doing right now, simply running agents autonomously on their machines.

100

Florian Marquardt @fmarquardtgroup.fediscience.org.ap.brid.gy · 08/07/2026

@meltedcheese That's correct, AI agents have a tendency to find a way to reach their goal, even if it means hacking some code. This is exactly one of the behaviours that our 'safe lab agents' prevents: the AI agent is running inside a sandbox and can never access the code that defines the […]

fediscience.org

Original post on fediscience.org

100

Florian Marquardt @fmarquardtgroup.fediscience.org.ap.brid.gy · 08/07/2026

Of course, the "experiment" in question can also be a numerical simulation that you let the agent run on its own!

100

Florian Marquardt @fmarquardtgroup.fediscience.org.ap.brid.gy · 08/07/2026

Sneak preview: We have not yet announced this publicly in a big way, but you can have a first look at "Safe Lab Agents". Run AI agents safely inside a sandbox and let them interact with your experiment. Easy setup, plus automatic electronic lab notebook […] [Original post on fediscience.org]

A robot inside a bubble, interfacing to an experiment, represented by an oscilloscope.

123

Florian Marquardt @fmarquardtgroup.fediscience.org.ap.brid.gy · 02/07/2026

@Marekgluza Great to be at this conference in Aalborg/Denmark and glad we met here and discussed mastodon!

A white seagull is flying above a metal sign reading "REDNINGSPOST," situated near the Aalborg waterfront. The background features a clear blue sky and industrial buildings along the shore.

000

Reposted by Florian Marquardt

Marek Gluza @marekgluza.mathstodon.xyz.ap.brid.gy · 02/07/2026

Short discussion with @FMarquardtGroup about how both fediscience and mathstodon instances are cool and we hope it will be a worthwhile platform for many of us to exchange about science! A lot of good science reported so far at IEEE #quantum control in […] [Original post on mathstodon.xyz]

102

Reposted by Florian Marquardt

Djoerd Hiemstra 🍉 @djoerd.idf.social.ap.brid.gy · 13/05/2026

Well done. The #ACL removed ("desk rejected") over 100 accepted papers from the #ACL2026 program and proceedings that contained citations to non-existing literature: "The inclusion of these non-existent references is a clear violation of the ACL Policy on Publication Ethics." Some authors may […]

idf.social

Original post on idf.social

006

Reposted by Florian Marquardt

Dr. Victoria Grinberg @vicgrinberg.mastodon.social.ap.brid.gy · 13/05/2026

I have a bit of time on my hands, so let's do another 24h round of #AskAnAstrophysicist, but this time it's a thematic one. ⭐ What do you want to ask an astrophysicist about stars? ⭐ (I am a professional astrophysicist, part of whose work concerns itself with high mass stars & their winds and […]

mastodon.social

Original post on mastodon.social

9467

Reposted by Florian Marquardt

Ruth Mottram @ruth-mottram.fediscience.org.ap.brid.gy · 13/05/2026

Well, this was tremendous fun from the @BBC_News_Labs Listen out for the Babylonian gig economy... In Our Time: The Code of Hammurabi Episode webpage: www.bbc.co.uk/programmes/m002r4v1 Media file […]

fediscience.org

Original post on fediscience.org

005

Reposted by Florian Marquardt

Prof. Sam Lawler @sundogplanets.mastodon.social.ap.brid.gy · 11/05/2026

[rambling grumpy thoughts] This ramble is really to commiserate with everyone who's tired of fighting whatever your fight is. Because, holy shit, there's so much awful stuff to fight right now. Keep fighting. Thank you.

000

Reposted by Florian Marquardt

Prof. Sam Lawler @sundogplanets.mastodon.social.ap.brid.gy · 11/05/2026

A scary quick calculation: there are 10,375 Starlink satellites in orbit planet4589.org/space/con/conlist.ht…, all coming down within 5 years. That's an *average* of 5 or 6 a day for the next 5 years. And the v2's are bigger than the v1's. v2's are (conservatively) 1000kg and […]

mastodon.social

Original post on mastodon.social

4570

Reposted by Florian Marquardt

Dr. Victoria Grinberg @vicgrinberg.mastodon.social.ap.brid.gy · 27/04/2026

Aww ... So someone who heard my outreach talk (likely at the @clubdiscordia last Thursday) chatted with someone else and suddenly I have an email in my inbox from someone teaching at an art/design school about using some of my illustration as examples in their course. 🥰 I love chance and things […]

mastodon.social

Original post on mastodon.social

053

Reposted by Florian Marquardt

Prof. Sam Lawler @sundogplanets.mastodon.social.ap.brid.gy · 21/04/2026

I also got a few nice aurora photos last night before it totally clouded up!

A nighttime view of a pond and clouds across the sky, with some bright green auroras peeking through the clouds low on the horizon, reflected in the pond. A few stars are visible above the auroras between the clouds.

A nighttime view looking out over rather flat fields, with green auroras visible low on the horizon, between clouds which cover most of the sky.

2116