> users will go to whoever has the best model Depends. You might want privacy, need low price in order to process big volumes, need no commercial restrictions,

> users will go to whoever has the best model Depends. You might want privacy, need low price in order to process big volumes, need no commercial restrictions, need a different tuning, or the task is easy enough and can be done by the smaller free model - why not? Why pay money, leak information, and get subjected to their rules? You will only use GPT-4 or 5 for that 10% of tasks that really require it. The future spells bad for OpenAI, there is less profit in the large and seldom used big models. For 90% of the tasks there is a "good enough" level, and we're approaching it, we don't need smarter models except rarely. Another concern for big model developers is data leaks - you can exfiltrate the skills of a large model by batch solving tasks. This works pretty well, you can make smaller models that are just as good as GPT-4 but on a single task. So you can do that if you need to call the API too many times - make your own free and libre model. I think the logical response in this situation would be to start working on AI-anti-malware, like filters for fake news and deceptive sites. It's gonna be a cat and mouse game from now on. Better to accept this situation and move on, we can't stop AI misuse completely, we'll have to manage it, and learn quickly.