I payfor the premium service models, several of them and many are excellent productivity boosters...BUT
Every so often after sometimes hours of good responses a model would spew some illogical babble quite smoothly like how using a certain scoring method the LLM invented you can achieve 102% score on a test.
If you have some unconscious biased the LLM may just reinforce it with a logical argument in favour that is plain false but you might not know because you're could genuinely not be even aware you're biased. (Unconscious bias)
Any response from any AI tool must be vetted and checked carefully. It is a tool, like any other and if you do not check your measurements and outputs etc you will get misleading and costly results.
The fact the Trump administration does shoddy homework (if any) is hardly surprising.