2 ms·
I wonder why OpenAI doesn’t run a “supervisor” ChatGPT instance in parallel with the regular one, that would step in before the regular instance utters somethin
by layer8 4y ago
I wonder why OpenAI doesn’t run a “supervisor” ChatGPT instance in parallel with the regular one, that would step in before the regular instance utters something the supervising instance deems unacceptable. Only because it would consume double the computational resources? Or because they believe they can still successfully maintain the “guardrails” without that?