There is no doubt that LLMs are advancing but now they have mainly transitioned to incremental refinements.
Why will this happen?
Learning for LLMs, which happens through data, is bound to plateau sooner or later, when most of unique patterns are learnt (and new info to be gained from data is marginal). There may not be a huge improvement unless there is some breakthrough like "attention is all you need" (2017) which made transformers dominant architecture.
Then why is GenAI now more popular than ever?
Companies are now focusing either on making GenAI applications useful for all kinds of IT work or trying to make infra powering GenAI/applications better, efficient and optimised.
Agentic coding, harnesses, RAG, voice assistants etc. are all great and popular applications of LLMs. Separating training and inference chips, KV cache optimisation etc are few examples of infra optimisation.
It's great time for people with ideas to be a builder