My unpopular opinion is that people adopting this type of machine learning will have some benefits and some speed-bumps and obstacles. As GPT iterates through a myriad of versions, improvements, restrictions, filters businesses will have to work around legal challenges, shadow biases, confidently incorrect interpretations and of course dependencies on a rate limited cloud service unless all future iterations become open source, self hosted and distributed and everyone can agree on what data-sets to ingest and what should be filtered/excluded for legal, accuracy and other reasons. Businesses taking this on early will need to be ready to quickly change course on the fly especially as new legislation catches up.
I think the biggest obstacle could be the lack of showing ones work when legal issues will arise. There isn't a "debug last answer" to get forensic data on how the answer what achieved. I am curious how a cranky judge may respond to "because the AI said so".
Another potential risk could be if shadow biases dynamic filtering, dynamic algorithms, dynamic tuning based on social, economic or political preferences of the ML operator, get too aggressive and people start to realize their financial decisions are being manipulated and impacted artificially by automation even more so than occurs today on social media platforms. I do not know how businesses or governments may react to this.