
Bioinformatician, Cellular Biologist, CSO @tychobio_ai @mathurindorel@fediscience.org @mathsrish.bsky.social
Berlin, Germany
Joined September 2016
- Tweets13.7K
- Following1.7K
- Followers434
- Likes83.3K
Mathurin Dorel retweeted
Replying to @Alan_Risal2
The primary goal shud be to learn causal models of regulation which wud enable perturb perturbations as a byproduct. Instead, the goal has become hill climbing on some eval metrics by any means. Inevitably incentives brittle shortcuts rather than making progress on real problem.
Mathurin Dorel retweeted
How pensions ate the state ft.trib.al/tjQgXFt
Mathurin Dorel retweeted
Un pays qui n'investit pas dans sa recherche est un pays qui se meure. On a pu trouver 1 000 000 de personnes pour signer une pétition contre CNEWS? on doit pouvoir en trouver 100 000 pour la recherche.
Sauvons le CNRS et la recherche publique. petitions.assemblee-national…
at long last we are choosing who will wield the Ring from the famous fantasy novel The Ring Will Corrupt Anyone Who Wields It
Mathurin Dorel retweeted
someone told me a few weeks ago that it’s not that hard to keep up with biotech startups, there’s only like 200 new ones a year but a far better use of brain space is to just multiply that by a hundred and learn all 20,000 of your proteins which won’t change and are real
Mathurin Dorel retweeted
Gemini 4 Neon: frontier-level intelligence with fast end-to-end task completion across complex software engineering, knowledge work, and science workflows.
Begins rolling out today in the Gemini API in @GoogleAIStudio, Gemini for business, for paid Google AI Pro and Ultra subscribers in Google @Antigravity, and for Google AI Pro and Ultra subscribers in the @GeminiApp.
Mathurin Dorel retweeted
Strongly believe building new wet lab infra in SF is investor catnip. There are many problems with biotech but the market for generating (quite sophisticated) datasets for cash is low on that list.
Cycle through the WPT (“WuXi/Pharmaron Test”):
- Is the assay something WuXi/Pharmaron can run?
- If so, is the team equipped to run it at the quality, price and speed of W/P? This is very difficult.
- If all of the above, can the assay change shape with minimal additional build out/time/expertise? Or are they accidentally building something with a fixed shape and the economics and pace of a platform biotech?
Think these differences in focus and capital allocation will show up clearly in growth over the next 6 months.
Mathurin Dorel retweeted
You gotta understand how much Ant and OAI would have to piss off enterprise customers for Microsoft of all companies to take this specific line in marketing
Mathurin Dorel retweeted
Turns out the French have unexpected leverage against the DOW. LOL. “The French can’t be bullied”.
Mathurin Dorel retweeted
Unfortunately, Russia is receiving dozens of ballistic missiles a month from North Korea. This is a grave threat not only to Ukraine. North Korea is honing its ballistic missiles by having them used in this war. This creates additional challenges for the entire region surrounding Korea. It is also a challenge for us in Europe.
If Russia has an ally that supplies it with ballistic missiles, then Ukraine also needs clear parity so that we can respond to the Russian threat on equal terms. Europe has missiles capable of carrying out effective strikes. Germany, France, the United Kingdom, Italy – we are discussing how countering Russia’s global banditry, supported by its ally in Pyongyang, must be backed by greater supplies of comparable weapons to Ukraine.
Responding in kind is fully justified. We also need to return to the discussion of destroying Russia’s ballistic missile launchers. The Western world has the technology to do this, and to bring Russia to peace sooner and guarantee security more reliably, we must have these technologies here in Ukraine.
From today’s evening address.
Mathurin Dorel retweeted
What we found:
- When different datasets profile same compounds in the same cell type or cell line, the size of gene expression changes agrees only weakly, while the direction of change is much better preserved
Mathurin Dorel retweeted
The fundamental problem with people who have never done frontier experimental biology work is that they say things like “solve biologists” without understanding what constrains biologists and what that means
Mathurin Dorel retweeted
Ship 42 returns to Starbase, Texas for the first-ever Starship tower catch during Flight 15! I have dreamt of creating this image for years... and today it became real.
Congrats, SpaceX team!
Mathurin Dorel retweeted
☢️🇫🇷🇬🇧 « Que tout État envisageant de menacer les intérêts vitaux de la France ou du Royaume-Uni sache qu’il serait alors confronté à la puissance combinée de nos deux forces. »
La France et le Royaume-Uni ont publié un communiqué conjoint d’une portée majeure et, il me semble, sans précédent, confirmant que désormais tout adversaire s’attaquant aux intérêts de vitaux de l’un ou de l’autre subirait des représailles massives des forces nucléaires des deux puissances nucléaires européennes.
La déclaration de Northwood initiale, divulguée le 10 juillet 2025, déclarait « nous ne voyons pas de situation dans laquelle les intérêts vitaux de la France ou du Royaume-Uni pourraient être menacés sans que les intérêts vitaux de l’autre ne soient également menacés. »
Il s’agit donc d’un langage beaucoup plus fort et d’un puissant signal de cohésion, tant à destination des adversaires potentiels que des alliés des deux pays en raison de la fameuse « dimension européenne » de nos intérêts vitaux.
Cette déclaration complexifie donc encore un peu plus les calculs de nos adversaires, en intriquant les différents enjeux européens.
A noter toutefois que chaque État garde la pleine souveraineté sur ses forces nucléaires et que la décision de leur engagement ne peut être autorisée que par leur autorité exécutive respective, le président de la République en France et le premier ministre au Royaume-Uni.
Enfin, la déclaration commune du 10 novembre ouvre pour la première fois la voie à l'utilisation des installations d'essais souveraines, au-delà de l'utilisation de l'infrastructure de radiographie éclaire EPURE que se partagent les deux États sur le site français de Valduc.
Ceci pourrait constituer une étape vers une plus grande autonomie du Royaume-Uni vis-à-vis des Etats-Unis dans le domaine du nucléaire militaire.
UK-French nuclear cooperation update! In case you missed it, the UK-French Nuclear Steering Group (set up under Northwood Declaration in the context of the Lancaster Treaties) met on 10 November. Identical statements: 1/x
gov.uk/government/news/uk-fr…
elysee.fr/en/emmanuel-macron…
Mathurin Dorel retweeted
Des fois je me dis qu'après la fresque du climat y'aurait peut être un truc à faire...
Mathurin Dorel retweeted
We just raised a $34M Series A at @orakldotbio, co-led by @NorrskenVC and @singular__.
AI has become remarkably good at designing molecules. Yet most cancer drugs still fail in clinical trials.
Here's why, and what we're doing about it 🧵
Mathurin Dorel retweeted
Let me know when you start doing spatial transcriptomics on your own brain.
Replying to @ramez
Where we're going, data shortages don't exist. If JEPA has teeth, spinning up your biology for unlimited testing on yourself, with no harm to your actual body is very within the realm of this technology. Current trials have to assess risk across a population. We are stepping into a realm where that's likely archaic. I'd imagine the question is, how much risk one individual wants to take, and defining the tolerances.
10 years does sound a little wild though.
Mathurin Dorel retweeted
Biopharma CEO "This presentation contains forward-looking statements, as a result of risks and uncertainties..."
AI/Tech CEO "We will cure all diseases in less than 10 years"
Mathurin Dorel retweeted
This got totally buried cause it came out the same day as Navier Stokes.
But I think this is super interesting evidence about what the true major source of algorithmic progress is.
Pretraining progress seems to be coming mostly from data improvements.
@who_is_jerbear and I pretrained combinations of year-representative open model recipes and data corpuses across 2019 to 2025 at various small scales.
Data improvements contributed 3.24x as many compute multipliers as model improvements did (12.0x vs 3.7x).
And the gains stack independently - a better dataset helps every architecture about equally, and vice versa.
Here are full results, plus what we think this means for the future of AI progress:
dwarkesh.com/p/pretraining-p…
Mathurin Dorel retweeted
Replying to @orukja
Learn how to think deeply about either the algorithms or the biology of a system. If you become a shallow generalist who relies on AI to handle the deep details, you will be outmoded and replaced. The latter is fashionable, but it's skipping building the real value of the PhD



































