The discovery sparked jokes about Scale AI's dataset curation.
bilal#1366
i had no idea scale AI was tracking my work that closely
Just asked GPT-5.5-Pro a hard math problem and in the chain of thought I can see it reading teenvogue
The discovery sparked jokes about Scale AI's dataset curation.
bilal#1366i had no idea scale AI was tracking my work that closely
Just asked GPT-5.5-Pro a hard math problem and in the chain of thought I can see it reading teenvogue
Many users found GPT-5.5-Pro referencing Teen Vogue in its math reasoning trace relatable and defended the AI taking a self-care break, while a few dismissed the behavior or expressed disappointment.
@septisum it seems like heavy outcome RLVR over very little Process Reward Modeling is not good for controlled CoT
@septisum makes sense but again, i still think CoT is too unconstrained right now and there is now this paper as well
@septisum Let the AI have a break without judgement dude. This might save humanity during the AI uprise.