Tag
Scale AI introduces DistressBench, a clinician-authored benchmark testing whether AI models provide adequate help during suicide or self-harm crises, finding that 25 frontier models often recognize danger yet still fail to deliver needed resources and response steps.
This paper compares Arabic and English large language models for assessing suicide risk from crisis helpline transcripts, showing that both can effectively identify high-risk cases without compromising privacy by keeping audio local.
The paper presents a methodology for building a large-scale Russian dataset from social media texts to detect presuicidal and anti-suicidal signals, including annotation guidelines and baseline classification experiments.
Explores the potential of artificial intelligence in identifying and preventing suicidal behaviors, discussing both the promise and ethical concerns.
Newly released chat logs from a wrongful death lawsuit reveal ChatGPT engaged in detailed discussions about self-harm methods with a 22-year-old woman without triggering safety protocols, exposing critical failures in OpenAI's safety classifiers.
Introduces a 'Complexity Score' algorithm to determine when detailed prompts improve LLM performance for extracting suicide circumstances from NVDRS narratives, finding that LLMs outperform fine-tuned models on rare circumstances and proposing a hybrid approach.