DeepSeek V4 got beaten by SimpleQA?
The real issue is the mismatch between high scores and actual capability. You can already see it from the base model’s score distribution: something feels off.
When a model optimizes for metrics but not intelligence, this is what happens.
Maybe it’s time for a rollback. Evil DeepSeek.
(Image translated from Chinese by grok)
@Xiphus1@AniTVOfficial Haven't read the manga and personally I'd like her to have feelings for him but could you develop what makes you say that in the manga?
Hey ! I love the tool, but was wondering if there is coherence in the generated maps (like if the placement « makes sense » and if a village will have a certain variety of buildings so there’s no « loopholes » like no water source or a big village having a tavern an things like that. To know if it’s suitable for ttrpg :)