Anthropic has developed three new defenses to combat data harvesting by AI distillers, reportedly reducing the amount of data collected by 90-99%. One of these defenses, known as 'preserved thinking,' is designed to enhance the multi-turn capabilities of their Claude agents. AI
IMPACT These defenses could significantly impact how AI models are trained and the privacy of user data.
RANK_REASON The item describes new technical defenses developed by a major AI lab for their models. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Mastodon — sigmoid.social →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →