There are several breakthroughs actually, every few weeks, inching closer and closer to the next generation of hardware for ML models with optoelectronic comput

There are several breakthroughs actually, every few weeks, inching closer and closer to the next generation of hardware for ML models with optoelectronic compute. This is just one of the most recent: https://news.mit.edu/2023/system-could-yield-more-powerful-efficient-llms-0822 https://news.mit.edu/2023/system-could-yield-more-powerful-e... Pretty much every single major tech university has their research/VC arm with a hand in photonics for AI. The initial rollout is around interconnect, but compute is moving along even faster than I'd expected, with significant steps forward occurring frequently. I'm skeptical we'll be seeing actual quantum computing any time soon for general purpose stuff due to error correction, but for the specific needs of training and running a neural network, it's pretty much the perfect marriage of hardware limitations with task specifications. So yes, this software trend is going to end up being much cheaper and efficient. By orders of magnitude leaps, not just at the creeping pace of Moore's law.