I feel that hallucination is not a deal breaker. I found it hallucinate less on topics with lots of data. For example, English grammar and python/css/html. > I

I feel that hallucination is not a deal breaker. I found it hallucinate less on topics with lots of data. For example, English grammar and python/css/html. > If a human has to validate the LLM output, that might take as much time (or more) as having done the task in the first place. I also disagree that validation is the always same amount of work as creation. In certain scenarios, I can verify the answers with less work. I usually ask ChatGPT to get an answer first and use google for validation.