cd/entity/Activation OraclesĀ· homeā€ŗ entitiesā€ŗ Activation Oracles
grep -l @activation oracles /news/*.json | wc -l ā†’ 1

Activation Oracles

mentions 1 type Person feed RSS

// recent coverage 1 mentions

02:54
2026-06-24
lesswrong.com
natural-language-processing

Can You Hide From a Natural Language Autoencoder?

Researchers stress-tested Natural Language Autoencoders (NLAs) by optimizing activation vectors to flip AV explanations while preserving model behavior, achieving an 81.4% flip rate with 99.6% label pā€¦

// co-occurs with top 5 entities
Qwen 1 Anthropic 1 Qwen2.5-7B-Instruct 1 Activation Verbalizer 1 Neural Chameleons 1
// topics top 5 topics
natural language processing 1 ai safety 1 ai research 1 machine learning 1 large language models 1