00:07:22	jasminec:	Wow, is this the Gabriele Sarti from this week’s reading? :0
00:07:25	jasminec:	Wow, superstar!!!
00:07:53	Grace Proebsting:	Reacted to "Wow, is this the Gab..." with 😮
00:44:57	jasminec:	Is it saying “does” and “equal” are more important?
00:47:00	jasminec:	Could we see you reason through more of these examples?

For instance, given the results from the previous attribution table (?) — what's an example of a concrete decision you might make? And how would you think through/design counterfactuals?
00:47:16	jasminec:	This is very helpful Gabriele :) !!
00:56:16	jasminec:	I’ll be honest, I’m not sure what to make of this
00:56:53	jasminec:	I’d expect it to be the animal who barks haha
00:57:31	jasminec:	Otherwise, why not choose a word like “eating” or another verb that some other entity could do.
00:59:22	jasminec:	Clever!
01:00:36	Arnab Sen Sharma:	maybe these attribution methods are just not good enough for modeling 2nd order effects? 
Even in the first case (wo contrastive) you kind of expect "dog" token to be super-important for "barking", right?
01:01:36	jasminec:	Reacted to "maybe these attribut..." with 👍
01:12:38	jasminec:	Just so I don’t get lost — what types of decisions can we make off of these results?
01:14:49	jasminec:	Oh, that sounds useful for some questions the power team was asking 🙂
01:19:44	jasminec:	Oh, that’s a great question!
01:21:30	Shuyi Lin:	My mic does not work
01:29:29	Emre Tapan:	thanks
01:32:41	jasminec:	Thanks gabriele!!!
01:33:03	jasminec:	Feels particularly/broadly useful in situations involving “high-stakes” decisions. 

Maybe where causal knowledge is much more needed than pure prediction 🙂
01:34:18	jasminec:	Thanks so much!