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 p…
Researchers introduced PROPEL, a framework that trains task generators for reinforcement learning by using a lightweight activation probe to predict solver pass rates, avoiding costly solver rollouts.…
A new study quantifies the rate at which undesirable behaviors transfer from teacher to student language models during distillation, a phenomenon known as subliminal learning. Researchers steered Llam…
Researchers introduced a paired fixed-content probe over 500 MMLU-Pro items to test how discourse-role labels such as Instruction:, Reference:, and Example: affect language model adoption of misleadin…