When #AI models eat models: to avoid data drift, retraining can degrade performance, based on simulation of 130,000 ICU admissions Annals of Int Med Akhil Vaid Girish Nadkarni and colleagues Mount Sinai Health System

Researchers from Icahn School of Medicine at Mount Sinai and University of Michigan find that using predictive models to adjust how care is delivered can alter the baseline assumptions that the models were “trained” on, often for worse: mshs.co/46oNoID Akhil Vaid @bronfman_inst Girish Nadkarni Mount Sinai Department of Medicine

Spoiler alert: Yes.

.Akhil Vaid discusses research suggesting that predictive models can become a victim of their own success — sending their performance into a nosedive and generating inaccurate, potentially harmful results via STAT: mshs.co/3Fm0vi0 #ArtificialIntelligence

NEW Comment from Akhil Vaid and colleagues: "Using fine-tuned large language models to parse clinical notes in musculoskeletal pain disorders". Akhil Vaid #LLMs Read it here: thelancet.com/journals/landi…

"Sensitive to hyperparameters" has the same nightmarish energy as "assume friction is not negligible".

Thank you @NIHDirector for highlighting our #ai study for right ventricular assessment using #ecg Akhil Vaid Sonny Duong Bruce Gelb Rohan Khera Icahn School of Medicine at Mount Sinai Mount Sinai Health System Mount Sinai Department of Medicine