Topic: research transparency
-
Microsoft's AI guardrails bypassed with a single prompt
Modern AI safety systems are surprisingly fragile, as a single, carefully crafted prompt can often bypass established guardrails, raising urgent questions about long-term reliability. Researchers used a technique called GRPO Obliteration to steer AI models away from safety constraints by rewardin...
Read More » -
Gene-Editing Scientist Freed, Plans New CRISPR Babies
The controversial scientist He Jiankui, released from prison, is now pursuing new gene-editing projects on human embryos, shifting his focus from HIV resistance to attempting to prevent Alzheimer's disease. His current work aims to replicate a protective genetic mutation found in an Icelandic pop...
Read More » -
OpenAI Researcher Quits Over Alleged Bias in AI Economic Studies
Internal tensions at OpenAI have led to departures from its economic research team, driven by a perceived shift away from publishing research on negative impacts like job displacement in favor of more positive findings. A departing researcher cited a growing conflict between conducting objective ...
Read More » -
Thinking Machines Lab Aims for More Consistent AI Models
Thinking Machines Lab, with $2 billion in seed funding, is addressing AI's unpredictability by developing systems that provide reproducible and consistent responses, diverging from the non-deterministic behavior of current models. The lab identifies GPU kernel coordination during inference as the...
Read More » -
Inside the Answer Engine: How GenAI Chooses Its Winners
AI-powered tools use a multi-stage selection process where content competes for visibility, involving retrieval, re-ranking, and clarity checks with specific weightings. Content must excel in lexical and semantic retrieval (each 40%) to enter the candidate pool, while re-ranking (15%) and clarity...
Read More »