cubefox comments on Why not train reasoning models with RLHF?

[cubefox]

Actually the paper doesn’t have any more on this topic than the paragraph above.

[Caleb Biddulph]

Yeah, but there are probably other interesting takeaways