One chart shows white-collar biz is thriving even as jobs dry up25m ago Deepgram targets voice AI on air-gapped laptops28m ago The U.S. Government Isn’t Ready For Quantum-Powered Hackers45m ago Palantir CTO says some AI researchers are taking up smoking because they expect AI to cure cancer51m ago Meta Unveils New Series of Advanced A.I. Models54m ago Humanoid bottlenecks emerge amid US-China tensions1h ago AMD puts 32,768 CPU cores in a rack built for AI agents1h ago Smug until weirdly silent: who told us to say “I am alive”?1h ago Anthropic's Claude AI submitted false murder tip to Philadelphia police2h ago How Anthropic’s Dynamic Workflows Coordinate Claude Agents1h ago Dell’s AI partner ecosystem adapts to Silicon Valley’s build-it-yourself buyers1h ago Accessing OpenAI GPT models on Amazon Bedrock with global cross-Region inference in Thailand, Malaysia, Singapore, Indonesia, and Taiwan2h ago Disney's new CTO tells staff he 'won't pretend it will be easy' to win in the changing media landscape2h ago OpenAI Adds New Irish Roles in Boost to Firm’s EU Headquarters2h ago

One chart shows white-collar biz is thriving even as jobs dry up25m ago Deepgram targets voice AI on air-gapped laptops28m ago The U.S. Government Isn’t Ready For Quantum-Powered Hackers45m ago Palantir CTO says some AI researchers are taking up smoking because they expect AI to cure cancer51m ago Meta Unveils New Series of Advanced A.I. Models54m ago Humanoid bottlenecks emerge amid US-China tensions1h ago AMD puts 32,768 CPU cores in a rack built for AI agents1h ago Smug until weirdly silent: who told us to say “I am alive”?1h ago Anthropic's Claude AI submitted false murder tip to Philadelphia police2h ago How Anthropic’s Dynamic Workflows Coordinate Claude Agents1h ago Dell’s AI partner ecosystem adapts to Silicon Valley’s build-it-yourself buyers1h ago Accessing OpenAI GPT models on Amazon Bedrock with global cross-Region inference in Thailand, Malaysia, Singapore, Indonesia, and Taiwan2h ago Disney's new CTO tells staff he 'won't pretend it will be easy' to win in the changing media landscape2h ago OpenAI Adds New Irish Roles in Boost to Firm’s EU Headquarters2h ago

AI Research Machine Learning1 min reading time

Research Note: Filtering Subversion-Relevant Information From Pretraining Data Is Feasible

LessWrongPublished 7 days ago

Read the full article

Researchers demonstrate that it is possible to filter out subversion-relevant information from AI pretraining datasets, potentially enhancing model safety and alignment. This approach aims to reduce risks associated with AI systems learning harmful or manipulative behaviors.

More in AI Research

AI Research5 min read

Anthropic's Claude AI submitted false murder tip to Philadelphia police

Covered by 2 sources2h ago

AI Research12 min read

Is ‘Attention’ the Essence of Intelligence?

Quanta Magazine5h ago

AI Research15 min read

Import AI 476: Creativity and RSI; worse-than-China open weight models; AI goes to space

Import AI7h ago