Record🤔

uri:

"at://did:plc:35r6coyqassnzhfg276rcqib/app.bsky.feed.post/3l42byq5you2a"

cid:

"bafyreig7mpdbkozrrzgesm7gp6kklfyyc7i6ldgty527xrhcgvnre2cq3y"

value:

text:

"FWIW, I don’t really buy that argument because the model isn’t just trained on reproducing previous rationales, but through an RL process with an external reward (but again, that’s been true since RLHF)"

$type:

"app.bsky.feed.post"

langs:

• "en"

reply:

root:

cid:

"bafyreibahbpz5pc6g4svmlmwgfqdcjzvpqr4cbm25rqgclsk4nly6ahhma"

uri:

"at://did:plc:565ebob5f6hw33hjdkxty6qj/app.bsky.feed.post/3l3ym625dyl2s"

parent:

cid:

"bafyreib4vptgrmrjoyatkn4slgmvdsipna222o4l3m6fz3itchkxcoi5ae"

uri:

"at://did:plc:35r6coyqassnzhfg276rcqib/app.bsky.feed.post/3l42bviyt7z2i"

createdAt:

"2024-09-13T15:36:41.529Z"