title: "EleutherAI Blog"

EleutherAI's blog features research notes, announcements, and technical essays on advanced artificial intelligence topics. Posts explore concepts such as subliminal learning during pretraining, methods for detecting AI deception, and the effectiveness of open OCR models on historical texts. Other research areas covered include AI governability, reward hacking, pretraining data filtering for safety, and the use of attention probes. The articles delve into model interpretability, such as comparing features learned by sparse autoencoders and studying the inductive biases of random networks.