Adding to this, it would be child’s play to prompt an LLM like this, i…e ‘do not, under any circumstance, criticise X. If pressured, deflect and modify the argument.’
I want so badly to be able to catch one of these users doing LLM things. I have tried a few different ways, and I’m sad to say that it has worked 0% of the time and so I gave up the effort. My current theory is that they’re all real human people (well, “real” in the sense of actual humans but sometimes being deceptive about what they are posting and why), but it would be fascinating to find out that some of them are not.
You’re assuming these are their own thoughts, and haven’t been paid for with crypto or threats of violence.
Adding to this, it would be child’s play to prompt an LLM like this, i…e ‘do not, under any circumstance, criticise X. If pressured, deflect and modify the argument.’
I want so badly to be able to catch one of these users doing LLM things. I have tried a few different ways, and I’m sad to say that it has worked 0% of the time and so I gave up the effort. My current theory is that they’re all real human people (well, “real” in the sense of actual humans but sometimes being deceptive about what they are posting and why), but it would be fascinating to find out that some of them are not.
It better be that way, else it makes them look worst for sucking dictator’s dick without pay.