RUMP SCIENCE

DEEP PRINCIPLE DROPS HYDROGEN-1.0: 753B OPEN-WEIGHT MODEL FOR SCIENTIFIC DISCOVERY

gnews-discovery · NOV 2

MERCK TEAMS WITH VARIATIONAL AI TO APPLY GENERATIVE AI TO DRUG DISCOVERY

gnews-discovery · NOV 3

[PAPER] STUDY: 92.5% of new peptide solubility dataset comes from a single historical programme

arxiv-science · OCT 31

[PAPER] STUDY: Allegro and MACE embeddings recover bonding descriptors with R² above 0.95 in linear probing...

arxiv-science · OCT 30

[PAPER] STUDY: New 'ROSE' AI functional beats modern neural rivals on chemistry benchmark

arxiv-science · OCT 31

[PAPER] STUDY: New AI workflow turns plain-language requests into fitted Landau energy models

arxiv-science · OCT 31

[PAPER] STUDY: More once-seen facts in training data, more discoveries by language models

arxiv-science · NOV 2

[PAPER] STUDY: Attribution reveals what single-cell foundation models see that performance metrics miss

arxiv-science · NOV 2

Korea launches $274mn supercomputer for AI drug discovery, advanced research

gnews-discovery · NOV 3

SCIENTISTS CAN NOW ASK AI TO DESIGN PROTEINS IN PLAIN LANGUAGE

gnews-models · NOV 3

[PAPER] Flat AI agent teams trailed a random-pick baseline in 20 searches...

arxiv-science · NOV 2

[PAPER] STUDY: AI research proposals match just 18% of human ideas, 26% of citations

arxiv-science · NOV 2

[PAPER] STUDY: AI-designed lattices hit 90% of theoretical stiffness limit at 0.4 relative density

arxiv-science · NOV 2

[PAPER] STUDY: Aggregate accuracy hides label-entropy confound in powder diffraction symmetry AI

arxiv-science · NOV 1

[PAPER] STUDY: Machine learning and DFT screen 1,356 boride surfaces as substrates for borophene

arxiv-science · OCT 31

[PAPER] STUDY: Learned models set record on turbulence benchmark, 39% below best published model

arxiv-science · OCT 31

[PAPER] AI SCIENTISTS OFTEN FAIL TO REPRODUCE THE RESULTS... 59 QUANTUM PAPERS, 11 AI SYSTEMS TESTED

arxiv-science · NOV 2

[PAPER] Agentic neuroimaging: CONFIRM cuts false support on controls from 23% to 0% when auditing an existing agent

arxiv-science · NOV 1

[PAPER] STUDY: HistoTx fuses tissue images with spatial transcriptomics in one transformer

arxiv-science · NOV 2

[PAPER] STUDY: AI gene-ranking gains come from recall of known genes, not the evidence it's shown

arxiv-science · NOV 1

[PAPER] STUDY: New AI model predicts full battery degradation curves, with errors below 2.5%

arxiv-science · NOV 2

[PAPER] Study: Disorder may boost battery-ion transport by way of the surrounding atoms, not the ion itself...

arxiv-science · NOV 2

[PAPER] STUDY: Training-free method beats AI at finding rare, highly functional sequences

arxiv-science · NOV 2

[PAPER] STUDY: Gas conditions alone failed to predict nanotube yield... growth curves help

arxiv-science · OCT 31

[PAPER] STUDY: AI fracture surrogates fail most in the final load steps, after the crack localizes

arxiv-science · NOV 2