Weekly reading
I highly recommend this article in the New Yorker about AI for mental health. The author is a hospitalist at Weill Cornell and a public policy expert. The article gives some interesting background about AI for mental health – did you know the first mental health chatbot was designed as satire, and its creator later became a vocal critic of AI? – and also an overview of the current AI/mental health landscape.
The authors of this Nature article from December 2022 developed a LLM from scratch from >82 billion words of clinical text, up from a previous maximum of 110 million parameters (ClinicalBERT, trained on the MIMIC-III database). Unlike MIMIC-III, which uses ICU data from a single site, this new model used notes from inpatient, outpatient, and emergency room visits in a heath system. They describe a 10% increase in accuracy for medical question answering and natural language inference. They even made the model, Gatortron, publicly available.
Fun Fact
The term “artificial intelligence” was first used by a Stanford professor in 1955
This Week’s Top Stories
University of Washington creates a medical data science institute
- The institute is focused on AI, machine learning, and healthcare
- Funded by the Provost with a $750,000 grant
- Headed by Sean Mooney, a PhD in biomedical informatics
First AI-designed COVID drug starts clinical trials
- Insilico, a Hong Kong-based company, will start trials in China
- Insilico leveraged AI in order to develop and test a molecule that can bind and inhibit the 3CLpro protease, the main protease in COVID
- Similar proteases exist in all coronaviruses
60% of patients are uncomfortable with their doctors using AI
- 57% say they are concerned that AI will make the personal interaction with their physicians worse
- 75% say they’re worried health care providers will move too fast implementing AI before understanding the risks for patients
Classic Deep Learning Resources
- Deep Learning by Ian Goodfellow
- What many have called the “Deep Learning Bible”
- GitHub lists of Deep Learning Resources
- Deep Learning Papers Reading Roadmap
- Awesome Deep Learning Papers
- These first two resources focus mostly on papers before 2017 due to the explosion in papers since that time
- Awesome Deep Vision
- Focuses on papers related to computer vision topics