Are LLM models approaching a saturation point?

LLM benchmarks are normalized between 0 and 100.<p>The main benchmarks alredy close to 100: - common sense reasoning (WinoGrande) - arithmetic (GSM8K) - multitasking (MMLU) - sentence completion (HellaSwag) - common sense reasoning &#x27;challenge&#x27; (ARC)<p>The only thing is if the Transformers architecture changes or if there are new benchmarks that measure the performance of models in new properties.<p>What&#x27;s next? Increasing performance and decreasing token cost has the potential to open up more complex use cases.<p>This will lead to the emergence of LLM processors and models will run entirely locally. This is a likely development scenario.<p>Any thoughts?