#AI4Science
LightNews LightNews
digital-discovery.rsc.org
Do materials foundation models learn actual physics, or just hallucinate structures? 🤔
SCALAR benchmark tracks what happens when models go from bulk crystals to 18,000-atom nanoparticles. Spoiler: high accuracy often hides severe physical inconsistencies #AI4Science
🔗 doi.org/10.1039/d6dd...
October 26, 2026 at 2:03 PM
"l’inactivation de DCTN1 pourrait-elle modifier la répartition du cholestérol dans les neurones, et ces changements pourraient-ils contribuer à la neurodégénérescence ?", ou comment l'IA guide la recherche... #AI4Science

openai.com/fr-FR/index/...
Des milliers d’agents d’IA pour révéler de nouvelles pistes dans les données biologiques
OpenAI, PerturbAI et Scripps Research utilisent des agents d’IA distribués pour révéler des pistes de recherche dans les données de millions de cellules cérébrales.
openai.com
October 29, 2026 at 8:57 AM
nnkennard.bsky.social
📢 AI use in peer review! 📢

Now that I have your attention, here’s a thread on our AIES paper about AI support for area chairs (the CS conference equivalent of a journal editor):

bsky.app/profile/nnke...

#peerreview #scienceofscience #ai4science
nnkennard.bsky.social Neha Kennard @nnkennard.bsky.social · 10d
Can the work of an area chair or scientific editor be automated? We interviewed 27 area chairs at a machine learning conference (ICLR) to find out what they *really* do and how (if?) AI can help. Findings and design implications from our #AIES2026 paper below!
October 25, 2026 at 2:42 AM
Science : le coût de la non reproductibilité était estimée à 28 milliards $/an en 2015 pour les essais précliniques... Est ce que l'idée de tests automatiques couplés à l'IA va améliorer les choses ? #AI4Science

journals.plos.org/plosbiology/...
The Economics of Reproducibility in Preclinical Research
This Perspective provides estimates of the rate of irreproducibility of preclinical research and its direct cost implications. It goes on to outline a framework for solutions and a plan for long-term ...
journals.plos.org
October 23, 2026 at 9:33 AM
aidailypost.com
Continuum‑LM shows how reusing dynamical operators in continuous‑depth models boosts exact‑match accuracy for AI4Science. Think transformer layers that learn once, work everywhere. Dive into the new scientific foundation model breakthrough! #ContinuumLM #ContinuousDepth #AI4Science

🔗
October 21, 2026 at 7:44 AM
Les géants de l'IA deviendront ils demain des géants des sciences ? Après Google DeepMind, c'est au tour d'OpenAI de rentrer dans l'arène, d'abord en math, et maintenant dans le domaine de la biologie... #AI4Science

openai.com/index/scalin...
Using thousands of AI agents to uncover new leads in biological data
See how OpenAI, PerturbAI, and Scripps Research used distributed AI agents to uncover promising research leads in data from millions of brain cells.
openai.com
October 22, 2026 at 11:54 AM
Après la révolution de l'IA basée sur les données (Alphafold), puis celle basée sur le raisonnement (Navier-Stokes), il est probable que la prochaine vienne de l'IA qui s'autoaméliore (RSI). Un papier sur le sujet. #AI4Science

arxiv.org/abs/2610.20431
SwarmRSI: Recursive Self-Improvement through Evolving Scientific Agent Swarm
Recursive self-improvement (RSI) in scientific research aims to improve agents' research capabilities through their own experimental experience. Scientific agent swarms can combine evidence and implem...
arxiv.org
October 21, 2026 at 12:38 PM
mattlease.bsky.social
Story on Sanjana Gautam's #facct2026 @facct.bsky.social #AI4Science paper on how Researchers Navigate Accountability, Transparency, & Trust with AI: ischool.utexas.edu/news/what-ar.... Sanjana is now at Microsoft

@texasscience.bsky.social
@utaustin.bsky.social
@msftresearch.bsky.social
#TexasAI
October 19, 2026 at 5:35 PM
bsc-cns.bsky.social
🏅Premi Marc Esteva Vivanco 2026 a la millor tesi doctoral de l’Associació Catalana d’IA (ACIA) per a l’investigador Luis Oliva.

Oliva és receptor d'una beca #AI4Science.

👏 Enhorabona, Luis!

@redpuntoes.bsky.social I #AI4S I #GeneraciónD I #PlanDeRecuperación #NextGenerationEU
October 14, 2026 at 9:54 AM
eosc-arena.bsky.social
That’s a wrap from the International AI Summit in Dublin 🇮🇪

A full day for EOSC-ARENA, FLUID-AI4EOSC and GenAI4Earth sharing ideas on trustworthy AI, FAIR data and research.

Thanks to everyone who stopped by our stand!

#EOSCARENA #AI4EOSC #AI4Science #EOSC
October 14, 2026 at 6:00 PM
eosc-arena.bsky.social
Halfway through the International AI Summit in Dublin 🇮🇪

EOSC-ARENA, FLUID-AI and GenAI4Earth are at the shared stand exploring how trustworthy AI, FAIR data and reusable models can support research.

Come find us this afternoon!

#EOSCARENA #AI4EOSC #AI4Science #EOSC
October 14, 2026 at 2:00 PM
xtalpi.bsky.social
Read more: prnewswire.com/news-release...

#AI4Science #MaterialsScience
XtalPi and Fangda Carbon Deploy Predictive AI to Optimize Cost and Material Efficiency in Graphite Electrode Manufacturing
The jointly developed AI model has passed comprehensive acceptance testing and is now operating within Fangda Carbon's production workflow. By forecasting...
prnewswire.com
October 8, 2026 at 11:10 AM
mattlease.bsky.social
@nsfsimonscosmicai.bsky.social
@texasscience.bsky.social
@utaustin.bsky.social
#TexasAI #AI4Science

Link to Thursday's talk details: amazon.science/conferences-...
October 7, 2026 at 7:17 PM
physicsteo.bsky.social
Bis bald, #AI4Science 2026🇩🇪.

Around all week to chat all #IntelligentDiscovery in the whole gamut of sciences⚛️🦠⚕️📜, and the challenges, and opportunities, of research in the agentic 🤖new world.

See somewhere on this map: https://teowaits.github.io/editors-and-beers/
☕🍺

#EditorsOnTour
October 5, 2026 at 5:21 AM
physicsteo.bsky.social
I expected delays on my way to Mainz🇩🇪, but expected them from DB. #Condor🦅 #AI4Science
October 4, 2026 at 7:49 PM
mkshurikov.bsky.social
🔥Excited to share our review on AI in organic synthesis! @wiley.com

We cover ready-to-use, actionable AI tools spanning the full experimental cycle—from synthesis planning to product verification.

Check it out!
doi.org/10.1002/tcr.70246

#Chemistry #CompChem #AI4Science #OrganicChemistry
October 1, 2026 at 7:26 PM
xtalpi.bsky.social
KQTD-126 was designed to inhibit TRK activity in the gut while limiting systemic exposure.

The next question is whether that profile translates in patients. The IND submission brings us one step closer to finding out.

Full release: www.prnewswire.com/news-release...

#AI4Science #DrugDiscovery
XtalPi Submits U.S. FDA IND Application for KQTD-126, a Potential First-in-Class, Gut-Restricted Pan-TRK Inhibitor for Chronic Intestinal Pain
This milestone marks the first IND submission from XtalPi's proprietary pipeline, recently unveiled in its interim results, validating the company's AI and...
www.prnewswire.com
September 30, 2026 at 2:44 AM
xtalpi.bsky.social
The question now is whether KQTD-126 can make chronic intestinal pain easier to live with.
The candidate’s IND submission is a step toward finding out.

Full release: t.co/QJu4sfJzFC

#AI4Science #DrugDiscovery
https://www.prnewswire.com/news-releases/xtalpi-submits-us-fda-ind-application-for-kqtd-126-a-potential-first-in-class-gut-restricted-pan-trk-inhibitor-for-chronic-intestinal-pain-302889891.html
t.co
September 29, 2026 at 2:52 AM
eosc-arena.bsky.social
AI is moving quickly into research, and EOSC-ARENA is bringing its 12 Use Cases together in a two-part webinar series to share what teams are learning and open the conversation to the wider research community.
📅 17 Nov + 16 Dec
🔗 arena.ai4eosc.eu/event/genera...
#EOSCARENA #AI4Science #OpenScience
September 28, 2026 at 5:00 PM
xtalpi.bsky.social
For researchers building synthesis planners: how do you turn a chemist’s correction into a lasting improvement in the system?

Read the preprint:
arxiv.org/abs/2609.05450

#Retrosynthesis #AI4Science
SureRoute: Toward a Hallucination-Free Self-Improving Platform for Retrosynthesis
AI models, including large language models, are increasingly integrated into scientific discovery workflows, yet they remain prone to hallucination. In experimental sciences, such errors translate dir...
arxiv.org
September 21, 2026 at 3:28 PM
xtalpi.bsky.social
For researchers building synthesis planners: how do you turn a chemist’s correction into a lasting improvement in the system?

Read the preprint: arxiv.org/abs/2609.05450

#Retrosynthesis #AI4Science
SureRoute: Toward a Hallucination-Free Self-Improving Platform for Retrosynthesis
AI models, including large language models, are increasingly integrated into scientific discovery workflows, yet they remain prone to hallucination. In experimental sciences, such errors translate dir...
arxiv.org
September 21, 2026 at 3:15 PM
dylan-ma.bsky.social
Everyone hates reading AI-generated content, yet everyone uses it. Collaborator often lean on AI to save time in joint project but obscures responsibility. Some thoughts on how to work efficiently on research collaboration in the AI era. #ai4science
machuangtao.github.io/blog/how-to-...
How to Efficiently Collaborate for a Research Project in the Era of AI?
Scientific collaboration is more than just sharing knowledge. It is an ongoing process of setting boundaries, sharing work, building trust, and giving credit. This article discusses how to protect you...
machuangtao.github.io
September 20, 2026 at 4:52 PM
vivienm.bsky.social
It's about Paper2Agent , also CellVoyager, VirtualLab, the conference AI4Science a conference with only AI papers and reviewers and more generally about the role AI agents play and the role human scientists must play.
September 17, 2026 at 3:33 PM
nan-bayesstat.bsky.social
AI4ScienceとSelective Inferenceの組み合わせは予想外
（講演を聴いてた）
September 17, 2026 at 3:30 AM
bsc-cns.bsky.social
🏅Premio Indra URSI 2026 al investigador del BSC Manuel Obelleiro-Liz en el marco del XLI Simposio Nacional de la Unión Científica Internacional de Radio.

Obelleiro es receptor de una beca #AI4Science.

👏Enhorabuena, Manuel!

#GeneraciónD I #PlanDeRecuperación #NextGenerationEU
September 15, 2026 at 8:23 AM