12,445 papers · continuously updated · last export: 10 Aug 2026livingmeta.ai
← Browse all papers
AI evidence extraction

MedGenesis: Toward a World Model for Autonomous Clinical and Translational Research

Xiao Hao, Nan Jiang, Tiancheng Zhang, Zhenfei Yin, Tao Gui, Zhongyue Zhang et al. · medRxiv · 2026

AI-generated evidence extraction, verified across multiple analytical personas. Not a substitute for the peer-reviewed original.

9/10
Relevance
0/4
Quality (LMQS)
C
Evidence
0
Citations
0.00
FWCI

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.64898/2026.06.14.26355612

Methodology & findings

Study design

Computational system development and benchmarking.

Sample

N = 1000000, 1 group

Primary method

World-model reasoning loop jointly updating Latent Hypothesis Space and Latent Action Space under expected information gain (EIG), uncertainty reduction (UR), and safety prior P(safe). Specific statistical software not mentioned in abstract.

Main result

MedGenesis demonstrated superior performance compared to frontier language models and biomedical AI systems. The paper reports that "MedGenesis outperformed frontier language models and biomedical AI systems while reducing hallucination" and "generated traceable outputs across meta-analysis, randomized controlled trials, real-world trajectories, case-control studies, and case reports, with one wet-lab-coupled run nominating a 3-hydroxybutyrate - neutrophil axis modulating antitumor immunity."

Reports effect sizes.

Research paradigm

Computational empiricism with AI-driven hypothesis generation and validation

Author conclusions

The authors conclude that "MedGenesis outperformed frontier language models and biomedical AI systems while reducing hallucination" and that the system's ability to generate outputs across multiple clinical evidence formats demonstrates that "hypothesis-to-evidence cycles" can be compressed "from years to hours, creating a continuous clinical discovery process."

Risk of bias

Not stated in abstract. Potential concerns include: benchmark selection bias, expert curation bias in ClinicalResBench, unclear representativeness of the 1 million patient observations, lack of explicit discussion of AI hallucination measurement methodology, and potential overfitting to benchmark tasks.

Open questions raised

  • The abstract does not explicitly identify future research directions or gaps. The work implicitly addresses fragmentation in clinical research tasks (evidence synthesis, mechanistic validation) but does not outline specific future work.
Data: not_statedCode: not_statedExtracted from: pdfAgreement 69%

Explore related topics

Meta-Analysis AutomationData AnalysisBenchmarking

Related papers

← Browse all papers