Tag
Research finds that GPT-4 can produce empty responses to null prompts while GPT-3.5 cannot, with cross-vendor studies confirming similar behavior in other models and an open-source tool introduced for controlling EOS token behavior.