Marcello Seri
LightNews LightNews
banner
mseri.me
Marcello Seri
@mseri.me
910 followers 760 following 590 posts
Associate prof at Bernoulli Institute, Groningen. Proud dad, mathematical physicist, #OCaml enthusiast, LEGO aficionado, amateur photographer, mushroom lover, fresh podcaster

#mathsky #physics #scicomm #ocaml #edusky

Also @[email protected]
Posts Replies Media Videos
Pinned
mseri.me
Marcello Seri @mseri.me · Feb 27
Hey, season 3 of It's Not Just Numbers is back! Tune in for the first two episodes of the year. Last in the row, an episode on study associations creators.spotify.com/pod/profile/...

#mathSky
S3E02 - Student associations, with Remco de Jong and Ilinca Antonache by It’s Not Just Numbers
In this episode we get to know Remco de Jong and Ilinca Antonache, board members of the mathematics and physics study association FMF. Together we get to understand what are study associations at Dutc...
creators.spotify.com
Reposted by Marcello Seri
barbarafantechi.bsky.social
It's not only Harvard. Universities worldwide have made conracts with AI companies, paying to get more tokens and/or tokens for more advanced LLMs. And yet, some people have "their" research stopped because they run out of tokens.

All of this in the current times of scarce and dwindling funding.
jedbrown.org Jed Brown @jedbrown.org · 7h
Oh. That's a heck of a subsidy.
September 16, 2026 at 9:54 PM
Reposted by Marcello Seri
ilpost.it
I criminali informatici non hanno violato direttamente i sistemi di Revolut: secondo informazioni date da loro stessi, si sarebbero serviti di almeno un indirizzo ufficiale di posta certificata delle forze dell’ordine italiane, violato in precedenza

ilpost.link/tzasnLnTbc
Revolut ha consegnato i dati di centinaia di clienti a criminali informatici
Hanno convinto la banca a farseli dare utilizzando una mail ufficiale delle forze dell’ordine italiane, violata in precedenza
ilpost.link
September 15, 2026 at 1:49 PM
Reposted by Marcello Seri
404media.co
EXCLUSIVE: Humans are reading ChatGPT users’ prompts to improve OpenAI’s models, and those chats can include sensitive, personal information, according to leaked internal documents and real prompts seen by 404 Media.
Inside ‘Project Lily’: The Humans Reading Your ChatGPT Chats
Humans are reading ChatGPT users’ prompts to improve OpenAI’s models, and those chats can include sensitive, personal information, according to leaked internal documents and real prompts seen by 404 M...
www.404media.co
September 14, 2026 at 2:26 PM
Reposted by Marcello Seri
mjb.mathstodon.xyz.ap.brid.gy
RE: https://hcommons.social/@mpe/117254767841427566

quite the achievement to get every living Fields medalist together to denounce something to which a handful of them have been actively contributing
mpe.hcommons.social.ap.brid.gy Martin Paul Eve @mpe.hcommons.social.ap.brid.gy · 5d
"The issues the mathematical community faces now are similar to issues that other scientific and creative professions are facing...: how to make sure that, as AI changes the way work is done, we do not lose sight of what that work was meant to achieve" https://mathandai.org/
Declaration — Math and AI
Read the declaration and add your name.
mathandai.org
September 11, 2026 at 11:22 PM
Reposted by Marcello Seri
msmathcomp.bsky.social
Here's your list of prompts for creating #MathArt throughout October. Happy #MathOber and looking forward to some beautiful #Art.
fractalkitty.com Fractal Kitty @fractalkitty.com · 6d
Mathober prompts have been posted and mathober website updated

www.fractalkitty.com/mathober-202...

and mathober.com

#mathober #mathober2026 #mathart
A list of days and prompts. In order: Perimeter of an Ellipse, Zebra Graph, Permutation List, Linear Recurrence, Inellipse, Connected Topology, Five Circles, Parabolic Rotation, K-Cyclic Graph, Simple Pole, Problem of Points, Reciprocal Prime Constant, Standard Map, Icositetragon, Irredundant, Isochronous Curve, Dimension Invariance, Biflecnodal Point, Orthic Triangle, Nice Vector Field, Prime Ring, Polylogarithm, Devil On Two Sticks, Multiperfect Number, Crossed Prism Graph, Unimodular Lattice, Indefinite Quadratic, Conic Double Point, Two Triangle Theorem, Spheroidal Wave Function, Existential Closure
September 11, 2026 at 6:55 AM
Reposted by Marcello Seri
philipcball.bsky.social
If you're wondering what has been going on in this domain, please do read this.
September 10, 2026 at 8:53 PM
Reposted by Marcello Seri
cailinmeister.bsky.social
Here is Holman and Bruner's wonderful paper on why this is so dangerous, and how the same strategy in the past earned a lot of money for pharmaceuticals at the cost of an estimated 100k lives
Experimentation by Industrial Selection | Philosophy of Science | Cambridge Core
Experimentation by Industrial Selection - Volume 84 Issue 5
www.cambridge.org
September 9, 2026 at 3:31 PM
Reposted by Marcello Seri
cailinmeister.bsky.social
By taking this money, social scientists allow Anthropic rather than the scientific community to decide what the agenda will be in AI research.
The Risks of Taking that Sweet AI Money
Anthropic is offering $200M to the funding-starved social sciences, but at what cost?
cailinoconnor.substack.com
September 9, 2026 at 3:30 PM
mseri.me
This surface visualizer is super cool: e-infinity.space/picture-book/

Ping @msmathcomp.bsky.social
A Topological Picture Book, Rendered
e-infinity.space
September 9, 2026 at 7:46 AM
mseri.me
Just a self-reminder for when the shady unethical behavior of large corporations seeking marketing stunts shadows my affection for mathematics: go on and read Francis Su "The Enduring Value of Math, in an Age of AI"

francissu.substack.com/p/the-enduri...
The Enduring Value of Math, in an Age of AI
Thoughts I’ll share at the start of my upper division math course
francissu.substack.com
September 9, 2026 at 7:45 AM
Reposted by Marcello Seri
707kat.bsky.social
- if OpenAI released its result in the way proposed I would go public with what happened.

(...)“Why would you ruin your career?” I replied that I am an academic, and asked why he thought going public would ruin my career.

(...) “If you don’t want me to be nice, then I don’t have to be nice.”
September 8, 2026 at 4:41 PM
Reposted by Marcello Seri
quantian.bsky.social
Dancing very very VERY close to saying “OpenAI stole our almost-complete proof of Navier-Stokes from internal access to our ChatGPT sessions because they were desperate to scoop Anthropic” which is absolutely 100% a believable thing OpenAI would do right now cims.nyu.edu/~tristanb/st...
September 8, 2026 at 7:13 AM
Reposted by Marcello Seri
dirkscheffers.bsky.social
Nieuwe column - English translation to follow.
ukrant.bsky.social UKrant @ukrant.bsky.social · 9d
Technologie is de sleutel tot succes, zei minister Letschert van Onderwijs tijdens de opening van het nieuwe Academisch Jaar. Columnist (en biotechnoloog) Dirk-Jan Scheffers waarschuwt voor al te hooggespannen verwachtingen. ukrant.nl/technologie-...
Technologie alleen gaat ons niet redden
Technologie is de sleutel tot succes, zei minister Letschert van Onderwijs tijdens de opening van het nieuwe Academisch Jaar. Columnist Dirk-Jan Scheffers waarschuwt voor zulke al te hooggespannen ver...
ukrant.nl
September 7, 2026 at 10:00 AM
Reposted by Marcello Seri
astrokatie.com
I sometimes send people this page: www.astrokatie.com/solicited-ad... which I hope is helpful. (See first paragraph for why I don’t provide individual mentorship to strangers.)
Solicited Advice — KATIE MACK, ASTROPHYSICIST
www.astrokatie.com
September 7, 2026 at 10:09 AM
Reposted by Marcello Seri
hetanshah.bsky.social
This is very interesting - a blog by a mathematician who was working on a proof of Fermat's Last Theorem on finding out that Anthropic has done it in 11 days. It is interesting because it does not concede the field, but says what it does / doesn't do
xenaproject.wordpress.com/2026/09/04/f...
FLT: Anthropic has beaten me to it
I guess technically it was revealed to the world by a coffee shop in Islington on Insta, but an hour later it was officially announced by Anthropic: one of their internal models, using the prove2.m…
xenaproject.wordpress.com
September 7, 2026 at 8:29 AM
Reposted by Marcello Seri
redpenblackpen.bsky.social
Todays substack on AI for paper writing - does it work? (spoiler: no, and you probably don't want it to either)

open.substack.com/pub/redpenbl...
August 3, 2026 at 7:07 PM
Reposted by Marcello Seri
redpenblackpen.bsky.social
Tag yourself with your current location!
A map in the style of Tolkein titled 'The Map of Manuscript Earth'. This is something I'm calling 'artography' - and it lists lots of geographic locations on the way to submitting a (scientific) manuscript.
August 7, 2026 at 8:43 PM
Reposted by Marcello Seri
redpenblackpen.bsky.social
Soooo.... did y'all get that this is calendar Tetris... or .... no?
redpenblackpen.bsky.social RedPen/BlackPen @redpenblackpen.bsky.social · 20d
I will say, it's soooo satisfying to see that entire day of meetings dissappear.

Though as you progress in your career the meetings start coming faster and faster. Eventually they're gonna overflow to Saturday and it's game over.
Comic with multiple panels showing a calendar scheduling grid where Monday is an entire red block with one tiny sliver. A voice bubble is saying 'Wow your Monday meeting schedule looks like hell!' - the second panel a blue meeting on Friday flips around and another voice says 'Nah- watch this'. The blue meeting block (L-shaped) moves to Thursday with a 'plunk' - voice says 'Whaa?!' - then moves to the Monday block with a 'shoop'. The entire Monday block dissappears 'Shiiiing!' and all the remaining meetings move to the left. Voice says 'Problem solved!'
September 7, 2026 at 5:01 AM
Reposted by Marcello Seri
markdingemanse.net
In case you'd like to try reading more than marketing claims and press coverage, this review article covers what makes it feel like LLMs do more than stringing together tokens: doi.org/10.5281/zeno...

Includes recent research showing the brittleness of 'reasoning' that experts here refer to
Although text-generating large language models have existed since 2018, it is only in the last few years that people have started to talk to them in earnest. The main reason is a series of innovations that enabled engineers to better tailor model output to human preferences and present it in a chat interface. Suddenly language models seemed to be much more than synthetic text extruding machine s (Bender and Hanna 2025) : they behaved as if following instructions, responded in ways that felt intuitive , and even seemed to have a mind of their own (Kockelman 2024). This interactivity, to a large degree enabled by unseen human labour, turned out to be tremendously compelling. My aim here is to demystify this phenomenon and so contribute some interactional building blocks to the sprawling edifice of critical AI literacies (Guest et al. 2026). Research that looks behind the curtains of benchmark results finds something quite different. One study by Anthropic researchers found that chain-of-thought traces often “systematically misrepresent the true reason for a model’s prediction” and identified a range of biasing features that made explanation-shaped model output “plausible yet systematically unfaithful” (Turpin et al. 2023, emphasis theirs) . Another study found that large language models, even if they deliver plausible output, “often arrive at correct answers through incorrect reasoning” (Nguyen et al. 2024). Even when tested on simple syllogistic reasoning , model performance is highly brittle and susceptible to stylistic perturbations (Ariyani et al. 2025) . In multiple choice question answering, the simple inclusion of an answer like “none of the other answers”, which requires real reasoning rather than stumbling around in latent space, leads to average model performance dropping from 66% correct to 31% correct on MMLU questions (Salido et al. 2026) . Meanwhile correct answers turn out to be hard to come by for truly new problems: when eight state -of-the-art models were tested on a set of brand new reasoning problems from the 2025 USA Math Olympiad, all performed dismally , and inspecting the “reasoning traces”, researchers found failures of logic, wrong assumptions, inability to choose between approaches, and miscalculations (Petrov et al. 2025).
September 7, 2026 at 5:51 AM
Reposted by Marcello Seri
mcnees.bsky.social
23 light hours, 47 light minutes, 20 light seconds and counting. Getting real close to a full twenty four hours for the one-way light time.
September 5, 2026 at 10:43 PM
Reposted by Marcello Seri
markriedl.bsky.social
Fixing the issue? Heavens no. But we’ll let you report it when it happens. Good luck out there
September 5, 2026 at 6:14 PM
Reposted by Marcello Seri
smbccomics.bsky.social
They actually say modulus, but journalists dumb it down.

COMIC ◆ www.smbc-comics.com/comic/imagin...
PATREON ◆ www.patreon.com/ZachWeinersm...
STORE ◆ smbc-store.myshopify.com
3-panel SMBC comic update. A man wearing glasses stands at a podium, declaring: "at last we've overthrown the economists! From now on the nation's finances shall be run by mathematicians!" Shortly after, the same mathematician declares: "as of today, money will be denominated in complex numbers. The public's long wait for an algebraically closed currency is at an end!" The next panel shows a news article headline: "ECONOMY CRASHES - Nation's money accidentally rotated 90 degrees, now imaginary. Mathematicians (quoted): "calm down, absolute value unchanged."
September 5, 2026 at 3:30 PM
Reposted by Marcello Seri
andrew.heiss.phd
You can use this UN-approved Equal Earth projection right now in #rstats #gis #databs
library(tidyverse) library(sf) library(rnaturalearth) world <- ne_countries() |> filter(sov_a3 != "ATA") ggplot(data = world) + geom_sf() + coord_sf(crs = "+proj=eqearth") + theme_void() A basic world map using the Equal Earth projection
September 5, 2026 at 6:13 AM
Reposted by Marcello Seri
euromathsoc.bsky.social
Looking for mathematical opportunities? 🔎

The EMS website lists job vacancies and upcoming conferences, workshops, schools and other events from across the mathematical community.

💼 Jobs: euromathsoc.org/jobs
📅 Events: euromathsoc.org/events
September 3, 2026 at 2:42 PM
Reposted by Marcello Seri
giuliofirenze.bsky.social
Pura potenza.

La spettacolare immagine che vedete mostra la contemporanea presenza di tre tempeste tropicali sul Pacifico centrale e orientale; due sono uragani maggiori (Lowell e Karina), la terza (Marie) diventerà un uragano nelle prossime ore. 1/5
September 2, 2026 at 6:06 PM