ChatGPT doesn't treat men and women equally

There's some underground AIs that seems promising, although they are being fed hand picked data, they also have a more focused use. When costs to train them lower in the near future, you'll hopefully see them pop up everywhere

That’s a bit of the thing- it isn’t even that it it is so particularly hard to create a decent “AI” as much that there is a lot of storage and you need the time and skilled labor to “teach” it. So in the ven diagram of places where the costs and effort are warranted, the resources and knowledge are available, and the application isn’t so critical that trusting it to a newer experimental technology with a questionable reliability aren’t a serious issue- there isn’t a lot of places where the three meet in the public space.
A problem with an early learning technology that learns slowly is that if you put the time and effort in to create a well educated machine, by the time you have a solid product, you could have put the resources into researching advancements to the technology and wether you did or not, someone did and you end up with a machine that is solid but essentially obsolete.

The modern crunch of sorts in tech- early adopters have always walked a bleeding edge where they shouldered costs and early adoption issues later adopters don’t. It has become the norm however that now early adopters are essentially BETA testers. Product goes to market in an unfinished state so that users can pay for the privilege of being early QA. It isn’t all cynical cost cutting- most products will gain feedback for changes or features in the early mass adoption stage you will miss in all but the longest most comprehensive BETA programs. Issues will be found and unforeseen use cases carved out. So it can get a better product to the consumer faster and cheaper. The main drive though is the market and technology move so fast. For many systems or products if you don’t get to the market before you’re ready you’ll be in the market too late.

In that sense many products aren’t even BETA for a product but more akin to a proof of concept- the first generation technology being largely a way to generate revenue and buzz and company name recognition or market share to finish development of the second or third generation product which is perhaps what we can call the “real” product. Machine learning is very similar. In the less visible sectors like big data and logistics and enterprise level stuff like that machine learning is off to pretty good strides. Things that are easily expressed and understood in numbers or have somewhat set or controlled variables and possibilities. Time and input are all that are needed after programming and implementation of a good design and solid use case. But many consumer facing applications are much more of a challenge and require all manner of context and layer on layer of assumed knowledge we often take for granted.

There are lots of potential applications for machine learning- from service or support like nursing or just abating loneliness to all manner of media and of course creating automation or increased functionality of cars and other machines that we use every day, but these tend to have steep curves in not only the sophistication of the “AI,” but in other practical aspects or social and legal complications. So a lot of the “cool” ideas that bring AI to the front of everyday life are at a level of complexity or effort and/or time to market with viable product that just make them sort of hard to fit into the modern consumer market cycle. Being the first in isn’t enough of someone else comes along a few months or a year or so later with a product that blows yours away.