Thank you so much for such a detailed comment.

I've many times talked in the past about LLM Modulo framework. LLMs on their own have already hit the wall.

But now with o1 and o3, we can barely call them LLMs, when most of the heavy loading is being done at test time reasoning which is more like MCTS.

So, I once again say, LLMs have reached the limits not AI.

In the below articles I've discussed symbolic and other type of architectures in full detail. It was months of efforts to write these ones, with some very cool hypotheses.

https://medium.com/aiguys/scale-wont-turn-llms-into-agi-or-superintelligence-75be01ed9471

https://medium.com/aiguys/the-hidden-limits-of-superintelligence-why-it-might-never-happen-45c78102142f