1. Combining Human Expertise with Artificial Intelligence: Experimental Evidence from Radiology by Agarwal at al. (2023). Although AI alone was more accurate in diagnoses than two-thirds of radiologists, supporting human experts with AI predictions did not uniformly increase diagnostic quality. The authors call the phenomenon “automation bias/neglect,” whereby clinicians tend to underestimate AI’s insights.
  2. What if Generative AI turned out to be a Dud? by Gary Marcus. Doubt by one of the most prominent AI researchers about the value that’s generated (and likely to be generated within the next years) by Generative AI. While the technology has found its use case in writing code or marketing copy, there are still a lot of hard to solve issues standing in the way of unlocking value for other fields.
  3. How Modernity Made Us Allergic by Theresa MacPhail for Noema. Article on possible causes behind the rise in allergies, both early and late in life.
  4. After 100 Years of Research, Autism Remains a Puzzle by Lina Zeldovich for Nautilus. A hundred years ago the Ukrainian psychologist Grunya Sukhareva first identified autism.
  5. How strong can you get without steroids? Strength standards for natural lifters by Menno Henselmans. Once steroids became popular it became impossible to tell what natural strength limits actually look like, so Menno collects data from the time before the rise of steroids.
  6. AI’s Political Spectrum at Exponential View. Summary of From Pretraining Data to Language Models to Downstream Tasks: Tracking the Trails of Political Biases Leading to Unfair NLP Models by Feng et al. (2023).