1 paper · 1 filter
Federico Torrielli, Gianluca Barmina, Andrea Blasi Núñez +4
Natural language autoencoders translate a language model's internal activations into readable explanations. Explaining every token position is costly. Which positions should an aud…