That's a very clickbait article title.
Remember when HTC released a blockchain phone? Barely. Because it was capitalizing on a fad.
AI/ML is a part of the future of computing, but it's a PART. Everything else matters too. In a few years, there will be good enough commodity ML models that none of these vendor AI models will be differentiators. Phones will need to be good enough to run these models fast enough, and we should see that on everything except the budget segment.
Meanwhile - and this is important - the performance of the rest of the phone matters. What good is amazing AI performance if games halfway through the phone's release cycle are sluggish, or network connectivity is compromised, or the battery life has halved?
I wanted to love the Pixel 8. I wanted to easily justify ditching my Z Flip 4, a phone that is just grating on everything except performance. But paying full price for a performance sidegrade is really difficult, and nothing in the AI/ML being shown right now is life changing - and I'm a software engineer who has done some AI work!
Moore's law isn't dead, but we are reaching the limits of blind parallelization. Right now, the things that provide the biggest actual performace increase are:
- Increasing the speed - both in clock speed and instructions-per-clock - of the fastest core in the processor
- Improving scheduling to make sure the fastest core is doing the hardest tasks
- Improving memory and storage bandwidth and latency - more cache, faster RAM, faster UFS storage, etc.
- Improving the GPU so it can consistently feed high refresh rate screens and complex games
- Making the software lighter (think Android's Project Butter a few years back)
The limitations of this round of AI will disappoint us. I fear we are approaching another AI/ML dark age if we don't temper expectations.