I love talking about AI. I have so many books on the subject, but rarely get to talk about it, so thanks for this indulgence!
I agree to an extent. Taking LLMs and simply making them bigger by throwing more data and computers at them is hitting a brick wall. The path we are on right now will likely never lead to true AGI. However, this depends as well on what we define as true AGI.
If we mean a machine with a human-like mind, consciousness, and genuine understanding, then yes, I think that is absolutely impossible. We might get to a stage where AI can mimic a human mind, but it would just be an illusion. An AI will never have consciousness or human emotions because those are a result of our unique biochemistry. This idea of AGI is, IMO, nonsense.
A more realistic definition and goal for AGI is looking at as software that can autonomously perform 80% to 90% of economically valuable white collar tasks.
With this definition the goal is shifted to functional automation by building a digital ecosystem of software agents that can autonomously run a company's data pipelines, write code, fix their own bugs, do 90% of office work etc. If this is the idea of AGI then that technology is feasible, although to get there we would need to probably move away from pure transformers (something like neuro-symbolic models)
If we can get to that form of AGI then we are talking about the automation of cognitive labor, which to put it lightly would be the most fundamental change in human history.