The current run-away hype-train with AI, it is inevitably going to crash into the mountain of broken promises, any day now.
https://www.theverge.com/2024/6/16/24179679/mcdonalds-ending-ai-chatbot-drive-thru-ordering-test-ibm
MacDonald stopping the use of LLMs for drive-thru order taking is maybe a sign for the changing tide. There is a lot of negatives with AI, the cost to train these models boggles the mind, and some places has put restrictions on data-centers due to water shortages.
Most of the demos has been manufactured to some degree, that Sora demo 
was not soras creation alone, check out the link here https://archive.md/5kWHu.
Other demos have had equal or more staging employed to make them seem more impressive than they might actually be.
Hearing From the Experts
I do not really trust most of what people involved in the big AI labs have to say on this subject, since they have a vested interest in making their models seem as close to breaking through the next barrier as possible. Whether that is true or not, why would investors throw money at them if they are far away from making new breakthroughs?
Another thing worth keeping in mind is that ML is not a discipline we started using in 2016, it is a little older than that.
https://www.techtarget.com/whatis/A-Timeline-of-Machine-Learning-History
If blockchain had to show disruption in its 13 year history, could we not apply the same thing and say that this has actually been a fair been underwhelming, we have spent 70 years, and all we got was this "auto-complete on steroids", with no actual reasoning capabilities.
Somehow this test which is one of my go-tos to test LLMs still somehow trips it up, if I am not mistaken alligator only contains 2 Ls. But according to the smart AI on you.com it does in fact have one hidden somewhere.
The really funny thing happens when you tell it that it is incorrect.
Amazingly it gets it right when it lists out the letters, but comes to different conclusion when it responds, how exactly this happens I am not sure.
I added a section from the previous picture to show that I didn't do anything funky between these two queries.
Though I have seen some AIs that is able to do this task correctly, not that this by itself is impressive or anything, I have the floor set so low that accomplishing this goal is not impressive, but failing it is a catastrophe.
If counting letters is beyond the AI how should I trust it to do something more complex for me?