Interesting trend suggested here: models that aren't smart enough to evaluate the fake paper (non-reasoners) are more likely to give (default to?) sycophantic responses.

Posters: xlr8harder. xlr8harder

2 likes recorded by the source corpus.