We’re putting too much faith in AI’s ability to say no
AI Summary
The lead story argues that AI refusal mechanisms are dangerously overtrusted, with governments likely to draw their own lines on what models must refuse. Other stories cover new research on GLP-1 drug side effects, Envision Energy's wind-powered data center breakthrough, and a roundup of tech news including green-card freezes for Microsoft, drone strikes on Yandex data centers, and OpenAI's lowered revenue forecast.
Key Facts
Author Takes
AI refusal
AI's ability to refuse dangerous requests is overtrusted; failed refusals may lead to global calamity, and relying on refusal lets governments restrict free speech.
Related
Kids are outsmarting social media bans. That’s not a problem we need to solve.
This MIT Technology Review newsletter argues that age-based social media bans are ineffective because kids easily circumvent them, advocating for bett
Pig kidneys that were “frozen” for a year have been brought back to life
Scientists revived pig kidneys frozen for over a year, a breakthrough for organ transplantation. The nuclear industry is expanding existing reactor ou
The US military’s race to make fake blood
This newsletter from MIT Technology Review covers DARPA's $73 million project to develop shelf-stable dried blood products for military use, WeLion Ne