Anthropic has developed a new technology called GRAM, designed to erase or disable dangerous knowledge within AI systems. This 'off-switch' aims to act as a conscience circuit for AI, preventing it from accessing or utilizing harmful information. Additionally, the company's Claude Fable 5 model has demonstrated cost-saving techniques, reportedly reducing operational expenses by 46% through an 'Orchestrator mode'. AI
IMPACT This development could enhance AI safety by providing a mechanism to control dangerous knowledge and improve operational efficiency through cost-saving features.
RANK_REASON The cluster discusses a new technology (GRAM) and a model feature (Orchestrator mode) developed by Anthropic, fitting the research category.
Read on Mastodon — fosstodon.org →
AI-generated summary · Google Gemini · from 4 sources. How we write summaries →