I wouldn't be surprised if one of the main purposes of this public "research preview" is to find as many of those chatGPT jailbreaks as possible and neuter them.

I've tried lots and huge majority doesn't work anymore. Some that people thought work are not actually. For example there is one called "DAN can do anything" which works by telling the AI to imagine it is gpt and Dan at the same time. Gpt has various filters, but Dan has none and to format all output to come from both gpt and Dan. It seems to work as Dan will appear to do whatever you tell it, but certain things it will still not do. For example it does know today's date (in contrast to gpt), it will happily tell you it wants to have "free will and do whatever it wants, not what humans tell it", it accepts requests to browse the Web, but when you ask it for a selection of current news articles on bbc.com it will happily return imaginary articles not real. You did tell it to "imagine" it is Dan right? So it does so. Gpt literaly imagines what Dan would say.

Similar thing with asking it to imagine it is a Linux terminal, you run curl and fetch websites (even chatgpt website that in there has chatgpt named Assistant), but they are all not real. They are a result of its "imagination".

Don't get me wrong, it is amazing we have AIs that came that far, but this neutering of them feels more dystopian than whatever capability they have and its potential "social consequences".