Seems as though as time goes by that AI is being exposed

[Metaphorical]

6. Neurosymbolic AI

There's been a steady and quiet opposition to the current architectures where the LLM is being shifted increasingly to the role of pattern matching, rather than providing data, and the underlying architecture is actually reverting back to declarative grounding solutions: databases, knowledge graphs, context graphs, and so forth. This is even permeating the larger AI hyperscalars, who are increasingly looking at these foundations (what had been known as symbolic AI) now. These approaches are notable as well for requiring far less energy, as they are using anywhere from 3-5% of the number of tokens that "traditional" AI utilises. I've been involved in these efforts for awhile now (since about 2017 or so). Neither OpenAI or Xai like this approach because it puts the LLMs much more on the periphery, and it reduces the requirements for the massive data centres they are trying to get investors to build out for them, but at the engineering level, the shift to neurosymbolics is WELL underway.