Nostr notes by Chloé Azencott

All what I've ever wanted, scientifically, in the last 20+ ...

2026-09-09T09:24:54Z

All what I've ever wanted, scientifically, in the last 20+ years, was to use maths and computer science to help advance biomedical knowledge. Machine learning for bioinformatics was a great (and, by then, niche) way to do that.

But now I've spent more than half of my career wading through the mud that started building up when statistical machine learning came out of its relative niche obscurity and was rebranded as artificial intelligence. It's becoming harder and harder not to drown, not to feel so very alone in both horrified by what's now sold as generative AI and believing in the usefulness of statistical models (not necessarily very computationally intensive, or rather, necessarily not very computationally intensive) for meaningful endeavors.

I sometimes wonder about giving it all up, but truth be told, I love my job. I love my research questions. I love digging and trying to make sense of a little piece of statistics here or of biology there. I love accompanying young researchers as they blossom and become experts in their own little corner of this super interesting field. I love listening to my colleagues explain how they build and use deep learning models that help answer actual biological questions. I love teaching and being thanked by students for introducing them to machine learning, even though some of them then go on to have brilliant careers at one of the Magnificent Seven, making me wonder how complicit I am.

Also, I wouldn't know what else to do. So on I go, trying to make the most of my academic freedom to keep doing unfashionable things and resisting the hype at a sweet spot.

In a partnership with Mistral AI, CNRS is deploying a ...

2025-12-03T08:27:31Z

In a partnership with Mistral AI, CNRS is deploying a locally-hosted LLM chatbot for its employees.

They provide some examples of tasks you can ask the chatbot: summarizing a long text; translating; and writing an outline for an article.

Tasks you'd want to delegate to an assistant because they're beyond you.

And then they decided it would be a great idea to "honor" Emmy Noether, one of the most influential women mathematicians ever, by naming this chatbot Emmy after her.

I cannot fully express how angry that makes me.

Can't be arsed to read something yourself? Ask good old Emmy to summarize it for you, that's right up her alley, she'll do it between proving two major results in algebra or mathematical physics, too bad she's only a website and can't make you a cup of coffee while she's at it.

https://alan-poc.dsi.cnrs.fr/accueil/index.html (in French)

I've been told recently that the EU is using Teams for ...

2025-09-19T08:26:24Z

I've been told recently that the EU is using Teams for everything, and that they've chosen it because they've been sold it's "private by design".

I'd be curious to have more details about this:
- is this true that every official business of the EU administration goes through Teams?
- is it true it's been chosen because they've been sold a "private by design" solution?
- is their Teams hosted on EU servers?
- what technical solutions could ensure that Teams be private (by design or not)?
- is there any reason to believe the entire written and video communication of EU officials isn't fed to the CIA and/or Copilot's training data?

(I'm not interested about mere anti-Teams/Microsoft sarcasm, as I can provide vast amounts of it myself unaided.)

"So, you're against using genAI agents in education? ...

2025-06-27T11:23:08Z

"So, you're against using genAI agents in education? What's your stance on calculators then? Gotcha here!"

My dude (why is it always a dude? No, don't answer me), your argument has more holes than a colander.

Calculators work. They do exactly what the people who want you to buy/use them claim they do. Not so with genAI chatbots.

Perhaps even more importantly, the existence of calculators does not negate the importance of teaching kids basic arithmetic. That is because what's important is not just the result, but also how you get to this result, and why you want this result in the first place. So even if genAI chatbots were failproof, I would still want students to be able to do the work themselves, because what matters is not so much the outcome as the process.

And, as using a calculator properly requires the ability to have some notion of what the result should look like, so as to reduce the very real possibility of a user error, using genAI assistants requires the ability to look critically at their output in order to fix all the errors (that, again, aren't due to user input so much as to their being error-prone by nature).

The initiative calling for the ban of "conversion ...

2025-05-17T07:24:41Z

The initiative calling for the ban of "conversion therapy" in the European Union has collected enough signatures (around 1 180 000 as I type) 🎉

But you still have till the end of today, May 17, to add your signature and support putting an end to hateful practices that have been described as torture by the United Nations.

Wouldn't that be a suitable thing to do on the International Day Against Homophobia, Transphobia and Biphobia?

https://eci.ec.europa.eu/043/public/

#IDAHOT2025 🌈

I'm so fed up* with articles about the widespread use of LLM ...

2025-05-09T11:11:58Z

I'm so fed up* with articles about the widespread use of LLM chatbots by university students that only interview students who use them intensively and teachers who have no idea how to encourage students to think by themselves.

Aren't there any stories out there of students who refuse to use them because they realize how detrimental to learning this can be (I'm less interested in those who refuse out of ethical/environmental/geopolitical concerns, not that those aren't valid) or of educators who have found how to effectively encourage students to do their own work, because the point is not the final product but the process?

(Yeah, my Introduction to Data Science course is starting in a few weeks and I'm not looking forward to the time we're going to waste on grading projects written by an algorithm.)

* by which I mean, excessively upset and having difficulty sleeping at night. The despair they induce has even been seeping into some of my dreams.

things might have been different had he stuck to cancer and ...

2025-03-26T20:47:15Z

In reply to nevent1q…gz08
_________________________

things might have been different had he stuck to cancer and forgone the whatever

To keep their federal funding, Columbia University agreed to ban ...

2025-03-22T07:14:03Z

To keep their federal funding, Columbia University agreed to ban masks, empower 36 campus police officers with new powers to arrest students, and appoint a senior vice provost with broad authority to oversee the department of Middle East, South Asian and African Studies as well as the Center for Palestine Studies.

https://www.wsj.com/us-news/education/columbia-university-trump-demands-federal-funding-e94d41ca

I don't have the words.

Public service announcement: you can use an Overleaf project like ...

2025-02-17T12:12:53Z

Public service announcement: you can use an Overleaf project like a git repo (Menu > Sync > Git).

You'll miss out on all the comments others write in the sidebar, but you'll be able to pretend it's like the old days, write in emacs, compile locally, work offline, and stop yelling at the user interface.

The last #ParisWiMLDS Meetup of 2024 is announced! Tuesday, ...

2024-11-14T14:29:21Z

The last #ParisWiMLDS Meetup of 2024 is announced!

Tuesday, December 10 at 6:45pm at Mines Paris PSL (right next to the Luxembourg garden).

Featuring
Meredith Whittaker [@Mer__edith](https://mastodon.world/@Mer__edith ) on AI, privacy, and the surveillance business model

and

Raphaëlle Rémy-Leleu
Raphaëlle Rémy-Leleu (npub1ru4…32a2) on ecofeminism vs techno police.

Followed by a cocktail sponsored by https://itn.dev/ and https://theallyance.one/

Mark your calendar, free your evening, and register for free at https://www.meetup.com/paris-women-in-machine-learning-data-science/events/304558096/?slug=paris-women-in-machine-learning-data-science&eventId=304558096

In 1985, Rob Wilsney explains that the early computer scientists ...

2024-10-11T17:30:39Z

In reply to nevent1q…9kpn
_________________________

In 1985, Rob Wilsney explains that the early computer scientists were mathematicians, and for them, mathematicians prove theorems and play chess, so they've been interested in building machines that prove theorems and play chess.

Isabelle shows several quotes illustrating the idea that men need to build machines to reproduce (self-engender) while women don't need to do computer science to have children.

#ParisWiMLDS

Isabelle reminds her of the original Turing test, which was an ...

2024-10-11T17:26:01Z

In reply to nevent1q…gdej
_________________________

Isabelle reminds her of the original Turing test, which was an imitation game, in which both a man and a woman are taskes to convince an interrogator that they are both a woman. If the computer can replace a man in convincing an interrogator that it's a woman, then it's intelligent.

#ParisWiMLDS

A bit of history: in 1948, Norbert Wiener says that the correct ...

2024-10-11T17:21:22Z

In reply to nevent1q…mh4r
_________________________

A bit of history: in 1948, Norbert Wiener says that the correct level to understand the world is not atoms or molecules but information.

This means that leaning and reproduction are not specific to animals any more.

This feeds a lot of fantasies.

This couple is presented as having created two beings, one of flesh and blood and one of bolts and nuts.

#ParisWiMLDS

Isabelle Collet @npub1v52…0dt6 is going to tell us about what ...

2024-10-11T17:08:44Z

Isabelle Collet Isabelle Collet (npub1v52…0dt6) is going to tell us about what makes computer science so unattractive to women but so attractive to men #ParisWiMLDS

Yes. As @npub1n5r…5g5f posted yesterday in another thread: ...

2024-10-08T09:05:35Z

In reply to nevent1q…pw8n
_________________________

Yes. As
Lutra (npub1n5r…5g5f) posted yesterday in another thread: https://mamot.fr/@juliegiovacchini/113270132112765249 she seems to have been the key person in this discovery, together with Rhonda Feinbaum (who, as far as I can tell, retrained as a genetic counselor after 18 years as a research assistant and lab manager)

I've tried to figure out if maybe she had left science, and ...

2024-10-07T18:10:42Z

In reply to nevent1q…tv8f
_________________________

I've tried to figure out if maybe she had left science, and no. She's a Senior Scientist in his lab. But only holds a BS? Why wasn't the first author of a Nobel-worthy paper given a PhD? https://www.umassmed.edu/ambroslab/LabMembersOnly/senior-scientist/

Lutra (npub1n5r…5g5f)

Too bad there was no woman in the vicinity of this Nobel-worthy ...

2024-10-07T17:37:09Z

Too bad there was no woman in the vicinity of this Nobel-worthy paper to share the award, really, one tries one's best, but it's so hard to find women who do science worthy of a #NobelPrize.

h/t [@juliegiovacchini](https://mamot.fr/@juliegiovacchini )

So my university is moving to Office 365, and we got an email ...

2024-10-07T11:36:14Z

So my university is moving to Office 365, and we got an email from the communication department promising (among other things) good security and accessibility, but now we've just received another email from support telling us to pretty please keep using our self-hosted cloud for sensitive data and for the love of all that's holy do not put it on Office 365.

I'd laugh if it wasn't so tragic...

Academia has a travel problem. Attending faraway conferences is ...

2024-10-01T08:20:45Z

Academia has a travel problem.

Attending faraway conferences is ecologically unsustainable, and creates a huge barrier to entry for people who are disabled, have care duties, or lack resources, among others.

One piece of the solution is to create local satellites of major conferences. This idea has been successfully implemented in Paris for #NeurIPS for a few years, and I'm happy to have joined the advising committee of NeurIPS@Paris (Dec 4&5 this year)

Check it out: https://neuripsinparis.github.io/neurips2024paris/

So I just gave this keynote on deep learning for clinical ...

2024-09-10T12:53:34Z

So I just gave this keynote on deep learning for clinical applications, right. Then some dude comes up to me, thanks me for my talk (always appreciated), and then tells me how nice he thinks it is that I'm handling the modeling myself. I'm a bit puzzled, I guess my PhD students have been handling it at least as much, but I just say thanks because socialization y'all, and then it hits me... does this guy think it's cute that I'm doing math? Is "professor of machine learning" misleading somehow?