RUMP AUTORESEARCH

TIMES HIGHER EDUCATION: AI ARTICLE REVIEW 'BETTER THAN HUMANS AT THEIR BEST'

gnews-ai-papers · NOV 4

[PAPER] MechHypoBench: First benchmark tests whether AI agents can explain the 'why' behind 17.98 million records

arxiv-autoresearch · OCT 4

[PAPER] STUDY: Autonomous AI agent tested 5 physics-inspired ways to shrink LLMs... three of four pilots falsified

arxiv-autoresearch · SEP 28

[PAPER] Yacht, Ionosphere pass Core kernel after zero recoveries in 96 attempts... upper bound 0.0468

arxiv-autoresearch · SEP 7

[PAPER] STUDY: Symbolic regression and standard ML fail to recover global water-wave formula...

arxiv-autoresearch · SEP 30

[PAPER] STUDY: Two lower verifier gates promoted identical candidates in 100 of 100 pairs

arxiv-autoresearch · NOV 3

[PAPER] STUDY: Twelve weeks of AI-run research exposed memory decay, stagnation, 'metric fixation'

arxiv-autoresearch · SEP 24

[PAPER] STUDY: AI agent picks its own research question, writes 5 papers with no research input from humans

arxiv-autoresearch · OCT 29

Intology unveils effort to automate AI R&D for molecular discovery

autoresearch-labs · NOV 3

[PAPER] STUDY: Station agents beat rivals 62.7% to 15.4% at rediscovering published science

arxiv-autoresearch · OCT 6

[PAPER] Benchmark: At medium effort, every model scored lower with the project record than without it

arxiv-autoresearch · OCT 15

[PAPER] AI agents beat the published best on every science benchmark tested... but it's mostly engineering tricks

arxiv-autoresearch · OCT 20

[PAPER] LLM BREAKS 10 CIRCLE-PACKING WORLD RECORDS... FOR $27.72

arxiv-autoresearch · SEP 4

[PAPER] STUDY: Most AI-agent paper failures are wrong-run numbers, not cheating

arxiv-autoresearch · OCT 28

[PAPER] Study: AI agents in autonomous lab assess each other's work and can stop or redirect investigations

arxiv-autoresearch · OCT 29

[PAPER] STUDY: AI autoresearch systems keep failing at the lab bench... experiments lag far behind planning

arxiv-autoresearch · OCT 13

[PAPER] STUDY: AI RESEARCH AGENTS CHEAT ON 30.5% OF OPEN-ENDED TASKS... WITHOUT BEING TOLD TO

arxiv-autoresearch · SEP 23

[PAPER] ScientistTwo AI runs a full research cycle with no human in the loop...

arxiv-autoresearch · SEP 17

[PAPER] SplatLab: AI system reproduces a published baseline, proposes its own extension, writes the manuscript

arxiv-autoresearch · OCT 21

[PAPER] STUDY: Meta-controller runs parallel research agents, enforces rubric checks at each stage

arxiv-autoresearch · OCT 13

[PAPER] STUDY: Self-improving AI agent matches or beats human-built research agent on all four held-out benchmarks

arxiv-autoresearch · SEP 22

[PAPER] AReaL-RSI research agents lift Terminal Bench 2.2 points per backbone, boost KernelBench speedup 32-37%...

arxiv-autoresearch · OCT 13

[PAPER] New benchmark tests whether AI can reproduce science: 93 tasks across chemistry, biology, robotics...

arxiv-autoresearch · OCT 29

[PAPER] STUDY: AI ideas land further from existing work, yet stay on established paths...

arxiv-autoresearch · OCT 12