@vikhyatk Trying to understand this weird behaviour- figured You guys are just SALTY @KanuGulati@PrismML@HessianFree
but you know what, Conway raised in March for web4 not even because of Underdog, not to even your little shitty model is only a negligible part of an App :)
@superalesha@PrismML Also we should start not reporting macro averages of benchmarks to inflate overall performance. Like grouping GSMK8k with hmmt etc. The 20% degradation on agentic tasks, e.g., are important for ppl to know