Imo a significant majority of frontier model interp would be possible with the ability to cache and add residual streams, even just at one layer. Though caching residual streams might enable some weight exfiltration if you can get a ton out? Seems like a massive pain though
Imo a significant majority of frontier model interp would be possible with the ability to cache and add residual streams, even just at one layer. Though caching residual streams might enable some weight exfiltration if you can get a ton out? Seems like a massive pain though