Massive Discrepancies in Playground Chat Function Response

Anybody else notice a huge difference between the way the chat function responses performed yesterday, and the way they (kinda didn’t) perform today? As of yesterday, GPT-3 was leading phrases with “I feel…:,” and “I like…,” and “You are….”

Today, we’re suddenly back at the cold and calculated leading phrases like “I think that…,” or “I understand that.”

Somebody messed with it and they shouldn’t have.

1 Like