Interesting, I’ve had the total opposite experience. GPT-4 is reasonable more often than not. I don’t find the “it’s sometimes wrong” argument very compelling because the same is true for 99% of other information sources. I’ve always had to use critical thinking when look for answers online anyway.