OpenAI has shown that these models at full power work great, so now they're trying to optimize for cost. I've gotten similar low accuracy responses from stuff it could handle a month ago.
It was kind of cringey when the model generated low accuracy nonsense the user detected that as "sentient." Come on
Idk man, I don't think gpts are sentient either, but I wouldn't laugh at that person's description; are we no longer sentient as we burble and fail to pronounce or string together words properly during a stroke?
I don't think being temporarily incomprehensible is a valid metric for sentience.
It was kind of cringey when the model generated low accuracy nonsense the user detected that as "sentient." Come on