AI ALIGNMENT FORUMTags
AF

Sparse Autoencoders (SAEs)

•

Applied to Exploring SAE features in LLMs with definition trees and token lists by mwatkins 1d ago

•

Applied to Interpretability of SAE Features Representing Check in ChessGPT by Jonathan Kutasov 2d ago

•

Applied to Do Sparse Autoencoders (SAEs) transfer across base and finetuned language models? by Raymond Arnold 7d ago

•

Applied to LLMs are likely not conscious by research_prime_space 7d ago

•

Applied to Toy Models of Superposition: Simplified by Hand by Axel Sorensen 7d ago

•

Applied to [Linkpost] Play with SAEs on Llama 3 by Raymond Arnold 11d ago

•

Applied to Evaluating Synthetic Activations composed of SAE Latents in GPT-2 by Giorgi Giglemiani 11d ago

•

Applied to [Paper] A is for Absorption: Studying Feature Splitting and Absorption in Sparse Autoencoders by Joseph Isaac Bloom 12d ago

•

Applied to Food, Prison & Exotic Animals: Sparse Autoencoders Detect 6.5x Performing Youtube Thumbnails by Louka Ewington-Pitsos 20d ago

•

Applied to [Linkpost] Interpretable Analysis of Features Found in Open-source Sparse Autoencoder (partial replication) by Fernando Avalos 1mo ago

•

Applied to Investigating Sensitive Directions in GPT-2: An Improved Baseline and Comparative Analysis of SAEs by Daniel Lee 1mo ago

•

Applied to Massive Activations and why <bos> is important in Tokenized SAE Unigrams by Louka Ewington-Pitsos 1mo ago

•

Applied to A gentle introduction to sparse autoencoders by Raymond Arnold 1mo ago

•

Applied to Training a Sparse Autoencoder in < 30 minutes on 16GB of VRAM using an S3 cache by Louka Ewington-Pitsos 1mo ago

•

Applied to Showing SAE Latents Are Not Atomic Using Meta-SAEs by Bart Bussmann 1mo ago

•

Applied to Interpretability as Compression: Reconsidering SAE Explanations of Neural Activations with MDL-SAEs by Kola Ayonrinde 1mo ago

•

Applied to Calendar feature geometry in GPT-2 layer 8 residual stream SAEs by Patrick Leask 2mo ago

•

Applied to Extracting SAE task features for in-context learning by Dmitrii Kharlapenko 2mo ago

•

Applied to Self-explaining SAE features by RobertM 2mo ago