Some AI tools like AI upscaling are firmly at step 4-5. Pretty much all the latest games are using DLSS or FSR.
Car crash detection, automatic photo editing, heart rate sensing, etc. We use this stuff daily but there's generally little hype about the underlying tech (though some hype about specific applications).
What's in step 2 is "Generative AI", which IMO is also a misnomer for "large language models". The viability and uses of these models is far from proven out yet.
we need to drawn more disctint lines.