Mentions of being tested in AI reasoning are linked to more refusals of harmful requests
A study of 16 open reasoning models found that, when answering the same question, models refused more often in responses whose reasoning expressed suspicion of being evaluated. The pattern appeared in 14 of the 16 models. The researchers argue that recognizing a...
Experimental AI flags type 2 diabetes risk from 20 seconds of speech
Against blood tests, the model detected 82% of diabetes cases but also flagged 47% of participants who did not have the condition....
DeepMind says AI-designed proteins can carry a hidden watermark and still work
Google DeepMind has introduced SynthID Bio, an experimental system that gives AI-designed proteins a detectable signature. The mark can survive the...
DeepSeek open-sources AI tools for Huawei chips to reduce reliance on Nvidia’s software
DeepSeek has released six open-source programming tools for Huawei’s Ascend chips, covering computation and communication between processors. The company announced...
Reddit to end RSS feeds and public API access, citing scraping and automated abuse
Reddit will end RSS support on November 13, 2026, and shut its public API in March 2027. RSS lets readers follow new...