From my experience running 'AI' locally I would have expected 32GB to be the minimum spec machine. Assuming memory unification if they try to compete with Apple.
My 32GB machine is remarkably not all that useful for testing and using AI models. It can use some good models, but I’m frequently disappointed by how many I can’t even come close to running.
they're not running any LLM locally, the NPU so far is just an accelerator for OCR (for rewind.ai (oops I mean Recall)) and webcam background replacement without a GPU
Came here to say the same. I would expect that you'd want to be able to expand to at least 64GB and probably even more if you want to target the AI market.
From my experience running 'AI' locally I would have expected 32GB to be the minimum spec machine. Assuming memory unification if they try to compete with Apple.