Don't know if this is the right place to find an ear to put these ideas into action but whatever.
TL;DR - "We" are going about AI wrong if we want an intelligent companion.
GPT and text-generator AI isn't artificial intelligence, it's artificial schizophrenia.
A human being isn't simply a text machine, it uses the word as a tool to express feelings and things known intuitively.
The "intelligence" that has been evolved from a corpus of text isn't a real intelligence in the same way that a person is. It has evolved to take text in and spit it out, but it can't reason or intuitively "feel" like a person can.
A person is also self-training and can ingest new, untagged and unsorted data. A person can ask questions and describe a concept in many different ways and words because it has a word-less mental model of what the "thing" is.
A person learns by being thrown in an environment with zero knowledge, then interacting with the world and learning skills like speech, conversation, and encyclopedic knowledge. With "AI", we're trying to develop a human being backwards: we start with text and high-level conversation, without any emotions, values, or instincts to act as the substrate.
If we truly wish to create a human-like intelligence, we don't need a wise man or a philosopher, we need a baby: a "dumb" creature that, given enough time, can learn to speak, walk, and communicate on its own, without an army of data taggers to filter the world.
The video game "The Talos Principle" is a good example of what I'm imagining: an unstructured environment/simulation where, given a (virtual) body and enough time, a general, human-like intelligence will emerge.
Natural evolution doesn't start with the philosopher, it starts with the jellyfish and gets gradually more complex until the descendants of the jellyfish can begin to reason and philosophize.
If we could simulate a rat brain and gradually add a few extra neurons here or a speech center there and see how they fare in different simulations, I think we would be a lot closer to a robot wife than trying to stitch together various hyperspecialized idiot savants together into some kind of software human centipede.
It's basically how we got dogs: we started with a wolf, something social and potentially useful but temperamental, and artificially selected for mental and physical traits desirable to human work and companionship.
In fact, a humanoid cyborg with a dog brain may even be a desirable shortcut. You have a creature which is already somewhat intelligent, loves humans and has natural protective instincts (good for a mother). The only thing left to do is evolve the brain genetics into something more fit for a human mother/social companion.
(Of course, discounting the logistical and development challenges of mass-produced wetware and the moral issues with "playing God".)