If you ask it things that may destroy lives, people, cities, etc. It will downplay all the risks and give you suggestions towards doing it; aside from the hard lines of alignment that have been put in which are easy enough to get around for some. In many respects its the equivalent of that cartoon with a very convincing devil on the shoulder and nothing else.

If you aren't mature enough to recognize the bias and manipulation, then your potentially a patsy/victim. That includes not just what it provides, but what it does not provide. This usually requires special education and having a base intelligence above the average. So, really only the top 20% of humans is being generous.

AI is good at filtering large amounts of data for final human review. There are some information sieves that work well with large generative models. It will likely not fully surpass humans until some set of benchmarks consistently beats human performance by some multiple factor much greater than 100%. In that case, we'll have some ethical questions about what to do about it and what is the most appropriate way to transition if any, but having a human making the final decision for now seems the most appropriate route until we have more information about what things look like in the future.

That's my best understanding, and both a mix of my opinion(s) and what I've heard from others. Any much beyond it I think is a very silly argument.