Who are these experts saying AI isn't a good R&D investment? Like I understand an LLM is just a token predictor but it's a damn useful one if you use it right, and all the other applications are still improving too (some examples in another comment of mine)
Treating it like an R&D investment isn't a problem - you only spend what you can afford to lose on R&D. In our case, we have no R&D budget and any money/time spent playing around with llm's leads directly to degradation of the core product and risks losing customers.