00:24:04	Aruna Sankaranarayanan:	How big is your dataset?
00:31:06	ch.wendler@northeastern.edu:	Do start from an automatically extracted circuit?
00:31:54	ch.wendler@northeastern.edu:	Why cannot we use ROME’s naive noising approach?
00:33:27	ch.wendler@northeastern.edu:	Causal tracing = ROME method?
00:34:09	ch.wendler@northeastern.edu:	These plots are amazing
00:41:53	ch.wendler@northeastern.edu:	+1
00:45:24	ch.wendler@northeastern.edu:	Would be really interesting to see whether this circuit is also active in the forward pass without the detailed instruction.
01:23:42	Aruna Sankaranarayanan:	Super cool stuff!