The problem has always been that every token is suspect.
It's the whole answer being correct that's the important thing, and if you compare GPT 3 vs where we are today only 5 years later the progress in accuracy, knowledge and intelligence is jaw dropping.
I have no idea what you're talking about because they still screw up in the exact same way as gpt3.
It's the whole answer being correct that's the important thing, and if you compare GPT 3 vs where we are today only 5 years later the progress in accuracy, knowledge and intelligence is jaw dropping.
I have no idea what you're talking about because they still screw up in the exact same way as gpt3.