emollick· @emollick · X·· 3 小时前AI 评分
SemiAnalysis 专家级工作基准评测的挑战
再次，@SemiAnalysis_ 在这里做了一件有价值的事，你也能看出专家级工作的基准评测有多么难以完成。我认为，在 ARC-AGI、GDPval、METR Long Tasks 等许多其他主要基准已经完全饱和的当下，这非常有用。
Again, @SemiAnalysis_ is doing something valuable here & you can see why benchmarks of expert-level work are incredibly hard to pull off. I think this is quite useful at a time where so many other major benchmarks (ARC-AGI, GDPval, METR Long Tasks) are completely saturated.
来源：emollick · x.com