The domain behind "AI" is so much more than just LLMs, and there is legitimately a lot of value to extract. There is meaningful measure of progress in the notion of having explicitly parallel processing units for co-processing, but this is just not it.
Even something as simple as sorting out through your photos based on visual information in them rather than filenames is yet an unsolved problem. We have good methods for local search, none of which gets shipped to the consumer.
There are legitimately important advancements to be made in day-to-day computing, covering still clunky and awkward experience of rudimentary tasks - but they are locked to academic Python and SaaS that does not care about retention. Even M$ and its 365 story cannot beat the utility that good old offline Office brought.
The real bubble is much, much larger than AI - and it starts to show. Code and businesses which are created just to be sold, with products that are designed to NOT succeed. The hype around big LLMs is just the cherry on top.
Lots of business leaders seem to think AI can replace human workers and make their company better.
Now ask them if they think hiring lower paid humans that struggle with basic logic and math, have no real world experience and can't tell fact from fiction would do the same. And if so, why haven't they been doing this already.
>why haven't they been doing this already.
Roger.
I think there's a pretty good number if they were going to do that, would have better utilized the regular employees they already had.
Plenty who would just want to make their company run cheaper. Which has been very common for something like centuries. For better or worse, sometimes good merit can be found there.
What about consumers?
For AI to fly off the shelf into consumer hands it would have to be cheaper than a regular PC too.
Not a special PC that's more exotic and expensive than a regular PC, nope, a special PC that's less exotic and less expensive than a regular PC. [0]
Mass-produced the same way it's always been done to make emerging features be taken in stride as the mainstream adopts them, at the same rate they become affordable at negligible additional cost to the predecessor commodity.
When the fundamental deficiency is because the AI is not yet good enough for regular PC's, it's not the most accurate response to blame the old PC for not being good enough for AI.
[0] It's quite possible that the most mainstream adoption of AI might be greatest in areas where the full numerical accuracy & precision of a PC is overkill anyway.