I wouldn't be so confident. Flash attention is a recent, significant improvement to training times and sure looks low-hanging now. I was not familiar with Edge

I wouldn't be so confident. Flash attention is a recent, significant improvement to training times and sure looks low-hanging now. I was not familiar with Edge AI, interesting concept. I feel like improving the efficiency of very large models is much more likely. The recent successes will lead to an even larger influx of $ in the short term --- this is an existential threat to Google, after all. We will see where things go!