Pity, this reward hacking bug by KernelBench was around July 2025 & it’s still wasting efforts/resources.
******************
Warning: This vulnerability led to artificial 18× speedups that were entirely fake - the actual computation performance was unchanged!
********************
Huge Thanks on hard work and bringing it to public @diyerxx . This is what actual research/Science is.
Complex Mistakes/Bugs are common but a straightforward bug like this in GT data is pretty embarrassing by @Apple research team. Totally change my PoV about their capabilities.
@volokuleshov do you believe everything you read on #NeurIPS posters?
Here are screenshots for Gemini, ChatGPT and Groq(Yes, even Groq detected the deception)
Hmm, I am not sure they hiding it or something, as I think they said something on the same line about using DeepSeek Architecture, IMO they just need to update @huggingface repo docs with more details. Am I right @MistralAI ?
✨Last minute shoppers get this perk 🎁 Order today to get before the New Year. Buy two get 35% off site wide effective today till end of the year. https://t.co/xt6odosa4l
So excited for tomorrow's drop of this bundle of sensory pleasure. 🧊 Stay tuned and get yours before it's gone. 🕯️
#frozenfig#scenscandle https://t.co/TAKaYhgRlf