My description was of me as a human user of ChatGPT fwiw, not the OpenAI API.
I had it again earlier:
Me: give me a bucket policy for write access from alb
CGPT: [waffle about IP ranges that is totally incorrect; then starts telling me ALB doesn't typically write to S3 because it's usually an intermediary between clients and backend services like EC2 instances or Lambda functions - it already knows from chat context I am using the latter]
Me: [whacks stop because it's rapidly getting out of hand] yes it does for access and connection logs
CGPT: To allow Application Load Balancer (ALB) to write access and connections logs to an S3 bucket, you need to set up a bucket policy that [waffle waffle waffle]
Me: [stop] yes I know that's what I asked for
CGPT: Here is an example of an S3 bucket policy [...]
Me: invalid principal [as far as I can tell, a complete hallucination]
CGPT: [tries again]
Me: yes I already tried that, valid policy but ALB still doesn't have permission
CGPT: [nonsense intensifies]
In the end I sorted it much quicker from AWS docs, which is sort of saying something, because I do often struggle with them. Thought I'd give ChatGPT a chance here but it really wasn't helpful.
Human users are charged by the number of messages, so longer responses are preferable because follow up questions use up your message allowance.
APIs are charged by token so shorter messages are preferable as you don’t pay for unnecessary tokens.