It's decent for scenarios where I need to look something up but don't have the most likely specific words for the concept, it seems to be able to consider synonyms and/or domain-specific jargon that I don't have a grasp of yet.
It's bad for scenarios where I need to accomplish something using a specific version of a library, language or platform. It can help to be very specific with your version numbers in your prompts, but at a certain point it doesn't seem to alleviate the effort of just reading version-specific documentation for whatever you're trying to do yourself.
It's worse than terrible at writing code, because it's great at writing code which looks correct and is thoroughly commented as though it understands what the code it's generating should be doing, but that code is often wrong in subtle ways.
Any ai falling from space and impacting on a poor developer, who was not looking up at the time, is bound to inflict an enormous fatal head injury.
The use of shitty AI will drive your blood pressure up and cause tons of git revert commits.
I had gemini pro do a ton of cleanup today. It did a good job but wiped out my limit.
Then i got to using gemini flash(my gpus are busy with another task) and it's so shit today. It's trash. It's absolutely no surprise to me that GPT 20B beats in basically every benchmark.