It might be advisable to test this, via prompt engineering, with the GPT-3/GPT-3.5 base models, i.e. with davinci and code-davinci-003. Otherwise it isn’t clear whether this behavior is influenced by some forms of SL/RL fine-tuning.
davinci
code-davinci-003
It might be advisable to test this, via prompt engineering, with the GPT-3/GPT-3.5 base models, i.e. with
davinciandcode-davinci-003. Otherwise it isn’t clear whether this behavior is influenced by some forms of SL/RL fine-tuning.