102 cases
1–24

Capital Region of Denmark mammography screening programme
verified evidence33.5% fewer radiologist screening reads

74.45% of 595 submitted code changes and 69.46% of 93,574 edits were LLM-generated.

Johns Hopkins Children's Center / Johns Hopkins Medicine
verified evidenceEye-exam completion: 22% → 100% in the control arm

Meta (Facebook)
verified evidenceRuns fewer than one-third of dependency-selected tests

Greater Houston primary stroke centers
verified evidenceMedian transfer time: 210 → 133 minutes

Swiss Plasma Center at EPFL and Google DeepMind
verified evidenceAcross the full 0.1-1.0 second interval, RMSE was 0.62 kA for plasma current and 0.75 cm for shape.

John Deere (Deere & Company)
mixed evidence59% average herbicide savings

The Permanente Medical Group
verified evidence3,442 physicians used the tool in 303,266 encounters; 968 used it in at least 100 encounters.

Klarna
mixed evidenceAverage resolution time: 11 → under 2 minutes

Twenty-eight South-East Queensland long-term care facilities
verified evidencePARO participants were significantly more verbally and visually engaged than plush-toy participants.

Rising Academies schools
verified evidenceTreatment growth was significantly higher with effect size 0.36 (p<0.001), described as roughly an extra year of learning.

TD Bank Group
mixed evidenceAverage preparation time: 15 hours → under 3 minutes

UCLA Health
verified evidenceNabla arm fell from 4:30 to 3:49; a 9.5% greater reduction than control.

University of California, San Francisco and UC Berkeley
verified evidenceMedian real-time decoding rate of 78.3 words per minute.

Waymo
verified evidence9 Waymo claims, an 88% reduction.

American Airlines
reported evidence20% shorter, about two minutes per flight and more than 11 taxi hours saved daily.

Deutsche Bahn
reported evidenceOne and a half hours with the digital E-Check; camera-gate traversal takes about five minutes.

Ocado Group / Ocado Retail
verified evidence227 units per labor hour in FY24, a 9.1% improvement.

Google DeepMind and EMBL-EBI
verified evidenceMore than 214 million predicted structures, covering nearly the full UniProt database

Audi Neckarsulm
reported evidenceAI analyzes about 1.5 million welds on 300 vehicles each shift

Be My Eyes
verified evidenceFourteen users described Be My AI support across personal and social contexts.

Lawrence Berkeley National Laboratory and University of California, Berkeley
verified evidence36 of 57 target compounds synthesized, a 63% success rate

Boston Public Schools
verified evidence50 buses removed in the first implementation year and about $5 million saved, with average ride time remaining about 23 minutes

Sixty-eight English secondary schools in the EEF Teacher Choices trial
verified evidence56.2 minutes in the ChatGPT group, 25.3 minutes or 31% lower.