Sabitlenmiş Tweet
As cheap compute access gets developed, and models start reaching the same benchmark limits, then efficiency of the models running inference is the priority. With similar low cost & efficiently performing models, the latency governs any advantage; and a compute node physically closer to customer/linked device or sitting in the local fog layer over 6G has the least latency.
Sustainable competitive advantages don't exist with models, they are becoming commodities, Meta releasing Llama models for free is a big hint. Also, several models are approaching similar benchmarks within months of each other.
As a commodity, compute efficiency is the main factor, so whoever can provide the cheapest per token cost is the winner. Though with no competitive advantage to limit entrants (even regulatory action is one political move away from reversal) and similar performance growth rates, the market will be diluted.
Also, I believe AI capital moves away from heavy centralized processing, and goes towards the edge and fog, because of latency. 6G turns the network into a distributed fog compute grid so models can be split and run across nearby nodes with almost no delay instead of everything going to some far away cloud.
English
