July 7, 2023: How smart is ChatGPT? Also, BMI is a poor indicator of health and what to do if you can't fall asleep. TGIF! —Andrea Gawrylewski, Chief Newsletter Editor | | | How smart is ChatGPT? Earlier this year, Eka Roivainen, an assessment psychologist at Oulu University Hospital in Finland, gave the chatbot the most commonly used IQ test, the Wechsler adult intelligence scale. Unsurprisingly, ChatGPT aced the verbal portion, scoring high on the vocabulary subtest, the similarities and information subtests, and general comprehension test. Its score? 155 on the verbal portion. For comparison, college-educated Americans score an average of about 113 on that part.
The caveat: Despite such a high IQ, ChatGPT is known to fail tasks that require humanlike reasoning or an understanding of the physical and social world. It's flummoxed by riddles and rather than using logical reasoning, it tries to rely on the vast database of information it has been trained on (useless when noodling puzzles).
What the experts say: "The very high Verbal IQ of ChatGPT combined with its amusing failures indicates there are aspects of intelligence that cannot be measured by IQ tests alone," writes Roivainen. And maybe it means we're not all out of a job just yet. | | | For as widely used as it is, the BMI (body mass index) scale is not the best indicator of health. Last month the American Medical Association recommended that doctors de-emphasize BMI because it can get weight-related health risks wrong, especially when used as the sole diagnostic tool. Some experts say it has become overused by physicians, and in the worst cases has been used to misidentify the weight status of people in many racial and ethnic minority groups.
Why this matters: Using BMI as an indicator of health can lead doctors to overlook underlying disease or spread stigma and shame around body size. Better ways to determine health include metabolic indicators (blood pressure, blood sugar, etc.) and muscle-to-fat ratio.
What the experts say: The AMA's statement "highlights the need for more research to understand aspects of the disease we don't understand very well, such as how health risks vary across demographics and the genetic, environmental and behavioral factors that increase the risk of developing obesity," says Francisco Lopez-Jimenez, a cardiologist at the Mayo Clinic. | | | Doctor AI Ok, ChatGPT is smart, but does it have the skills to diagnose your medical problem? Senior editors Tanya Lewis and Josh Fischman examine the strengths and weaknesses of using chatbots to diagnose health concerns on the latest episode of Science, Quickly. Listen now or subscribe wherever you get your podcasts. | | | ICYMI (Our most-read stories of the week) | | | • Beyond Ozempic: New Obesity Drugs Could Be Cheaper and Even More Effective. | 4 min read | | | • Watch Baby Octopuses Hatch from a Surprising Deep-Sea Nursery. | 5 min read | | | • Lab-Grown Meat Approved for Sale: What You Need to Know. | 5 min read | | | Have you had a chance to play around with ChatGPT yet? I've had some interesting conversations with it so far--its book recommendations were pretty good and it even came up with some alternative names for this newsletter! But for now I'm sticking with Today in Science. | Tell me about your ChatGPT interactions and any other feedback you may have: newsletters@sciam.com. Have a great weekend! | —Andrea Gawrylewski, Chief Newsletter Editor | | Subscribe to this and all of our newsletters . | | | | Scientific American
One New York Plaza, New York, NY, 10004 | | | | | | Support our mission, subscribe to Scientific American | | | | | | | | | |
