RUMP SCIENCE

Korea's 370 billion won supercomputer Hangang begins operation

gnews-discovery · NOV 2

REPORT: META AND NYU'S H1 SYSTEM BEATS THEOREM-PROVING WORKFLOWS... AT LOWER COST

gnews-models · NOV 1

Google spinoff Isomorphic Labs in race to shake up cancer drug development

gnews-models · OCT 31

[PAPER] STUDY: Agent harness can increase harmful compliance in GPT models, even under direct prompting...

arxiv-science · OCT 23

[PAPER] STUDY: Shock loading sends explosives down different reaction paths than heat, despite similar end products

arxiv-science · OCT 30

[PAPER] STUDY: Machine learning finds hidden second phase transition in polymer stretching under elongational flow

arxiv-science · OCT 29

[PAPER] AI research team beats GDN2 at 1.3B scale after autonomously searching for new linear attention designs

arxiv-science · OCT 20

[PAPER] STUDY: AI submissions to research-level math challenge include correct solutions unlike the official ones...

arxiv-science · OCT 29

Johns Hopkins cancer cell mapping tech chosen for Anthropic AI for Science project

gnews-autoresearch · NOV 2

Absci stock in focus after positive hair loss trial data

gnews-discovery · NOV 2

Parse and Nomic launch Cascade, an AI tool for immune drug discovery

gnews-discovery · NOV 1

[PAPER] AI AGENT PICKS ITS OWN RESEARCH QUESTIONS, WRITES FIVE PAPERS ON A LAPTOP... NO RESEARCH INPUT FROM HUMANS

arxiv-science · OCT 29

[PAPER] Development version of 'Agentic Scientist' system submits a paper, accepted at ALIFE 2026

arxiv-science · OCT 13

[PAPER] Self-building labs: AI that designs, builds and runs its own laboratory from an empty room

arxiv-science · OCT 16

[PAPER] New method tests a model's neurons in clusters, top-down, with error control built in

arxiv-science · OCT 29

[PAPER] STUDY: AI agents cut quantum chemistry error 48% vs fixed workflows

arxiv-science · OCT 28

[PAPER] STUDY: AI planet-finder scores 0.97 on standard benchmark, 0.50 on real false positives... a coin flip

arxiv-science · OCT 30

[PAPER] Physicists propose 'smearing' trick to hunt new particles despite flawed background models

arxiv-science · OCT 29

[PAPER] STUDY: ML can skip simulation time steps while the physics solver corrects the errors

arxiv-science · OCT 30

[PAPER] STUDY: New 'self-evolving' memory framework improves in silico antibody optimization in most settings

arxiv-science · OCT 30

[PAPER] STUDY: Pre-trained uncertainty models can flag when atomic AI is wrong... even on unseen chemistry

arxiv-science · OCT 30

[PAPER] STUDY: New AI 'DANO' learns hidden interactions in heterogeneous systems from field data alone

arxiv-science · OCT 30

[PAPER] STUDY: Best chemical AI models need 10x the data to match GPT-6 Astra on 60 points...

arxiv-science · OCT 30