We should be pretty confident that that one extrapolation plot in the GPT-4 tech report is a mu-transfer replication as well. A citation of the paper appears in their bibliography but I didn't see a usage of it in the text which is funny.
(OpenAI collaborated on the last Tensor Programs paper but not all of them, so maybe it counts as 80% of a replication.)
We should be pretty confident that that one extrapolation plot in the GPT-4 tech report is a mu-transfer replication as well. A citation of the paper appears in their bibliography but I didn't see a usage of it in the text which is funny.
(OpenAI collaborated on the last Tensor Programs paper but not all of them, so maybe it counts as 80% of a replication.)