Illogical Chatbots, Bioengineers’ Visions, & Adolescents’ Mental Health (Patreon)

Want Good Maths? Think Like a Trekkie

A team from the cloud computing company VMware discovered that the reasoning skills of Large Language Models (LLMs) like ChatGPT depend on which mood the prompt suggests. In a recent preprint, they report testing ChatGPT, Bard, Meta’s Llama, and Microsoft Copilot. They found that asking the models to think positively often (though not always) enhanced model performance. Perhaps most amusingly, the study uncovered that asking chatbots to consider a question in the context of Star Trek improved mathematical reasoning. Overall, they say that the quality of model output is unreliable. “Highly illogical,” Spock might have said. Paper here.

What’s Next for Biomedical Engineering?

The IEEE has released a white paper outlining the prospects of biomedical engineering. The engineers paint a bright future: Among other things, they foresee that doctors will create full-body digital twins of their patients that will make disease prediction more accurate, the increasingly common use of wearable sensors to track physical health and brain functions, the use of electronics to augment both physical and neural function, and new technologies leading to vaccines against cancer and opioid addiction. Paper here.

NAS Chimes In on Social Media Impact on Adolescent Mental Health

A recent report from the U.S.-American National Academy of Sciences (NAS) concludes that there is no evidence linking social media use to adolescent mental health, with the exception of those who experienced cyberbullying, cyber-stalking, and cyber-harassment. The NAS report stresses that social media also has benefits and that generalising conclusions are difficult to draw because the effects of social media depend both on the platform and how individuals use it. This agrees with my recent summary of the literature which you can find in this video.

Read the full report here, and a summary here.