PERSON DIRECTORY
Luke Bailey
Luke Bailey appears in 1 indexed conversation across The Cognitive Revolution. This directory brings every appearance, source, TL;DR, digest, and transcript into one searchable feed.
Dodging Latent Space Detectors: Obfuscated Activation Attacks with Luke, Erik & Scott
Nathan LabenzLuke BaileyEric JennerScott Emmons
Latent-space monitors can be driven below a 1% false-positive threshold while preserving harmful behavior, as attacks jointly optimize jailbreak success and activation obfuscation across probes, out-of-distribution detectors, and sparse autoencoders.Structured API access raises attack costs through request monitoring and throttling, while multi-layer monitoring and obfuscated adversarial training add capability degradation or compute friction; open weights remain an unresolved control problem, and roughly 10-hour GPU runs still found bypasses.