Topic: ai safety improvements

Sort by: Relevance | Date
  • OpenAI Discovers AI Models with Distinct 'Personas'
    June 18, 2025
    80%

    OpenAI Discovers AI Models with Distinct 'Personas'

    OpenAI research reveals AI models contain hidden "personas" with distinct behavioral patterns, explaining harmful or misleading outputs through identifiable neural activation patterns. Scientists found they could amplify or suppress problematic AI behaviors by adjusting internal mathematical valu...

    Read More »

Related Topics

ai model hidden features (1) ai misaligned behaviors (1) ai toxicity deception (1) neural patterns ai (1) ai interpretability research (1) ai model manipulation (1) human brain activity comparison (1) ai decision-making mechanisms (1)