arXivCeleste De Schamphelaere, Juan J. Vazquez, Ari Spiesberger, Tomáš Gavenčiak, Gavin Leech, Euan Ong, Jiahai FengThu, Nov 5, 2026, 10:54 AM PST
score 16.9
AI models that generate text to explain other AI models' internal directions
Original: Towards interpreting any residual stream direction by scaling activation inversion
Source: arxiv.org ↗
Writing ELI5 summary…