it doesnt count when you train on benchmarks
Wake up babe, Nvidia just casually dropped a 70B model that beats GPT-4o and Claude 3.5 Sonnet.
And it's fucking open source.
Try it: https://build.nvidia.com/nvidia/llama-3_1-nemotron-70b-instruct
Or
https://huggingface.co/chat/models/nvidia/Llama-3.1-Nemotron-70B-Instruct-HF
Or
MLX (image attached) 
Discussion
weird take, but ok lol
He's not wrong, though... π
The cursor.ai team have a good explanation of the nuance here:
Ahh I hadnβt thought about it from the coding aspect lol