But for real, some of these investors are just wasting money on training, once models are out retraining them to perform specific tasks is cheap, and companies will want to do it on their own because of privacy etc.
Maybe OpenAI and Google are safe for now because they keep everything closed, and maybe you really need a trillion paramaters for a general-purpose model, but I think for most things smaller ones will be enough.
All the companies spun up around OpenAI's API will probably fail hard, since they don't really offer much that others can't integrate on their own.
Any positions around prompting and integrating LMs etc. will be filled by people with background in NLP anyway, and the excess will be shaved off.
Cybersecurity will have a blast though, since LMs introduce a new class of vulnerability, it's super difficult to defend against and requires compromises.
I still feel hesitant to try ChatGPT only in the sense that it feels like a data collection nightmare. I'm fine using a throwaway email, but there's no way I'm putting my actual phone number there.