4 ms·
I have a theory that the more you use ChatGPT, the worse it becomes due to silent rate limiting - farming the work out to smaller quantized versions if you ask
by d13 2y ago
I have a theory that the more you use ChatGPT, the worse it becomes due to silent rate limiting - farming the work out to smaller quantized versions if you ask it a lot of questions. I’d like to see if the results of these tests are the same if you only ask one question per day.
- slushy-chivalry 2y agothat's an interesting hypothesis, I suppose one can make N calls to the API and look if the distribution of wrong answers is skewed towards the later portion of the API calls
- OxfordOutlander 2y agoI wouldnt expect this from the API, because each token is the same revenue for OAI. With chatGPT however, you pay a flat rate, so every incremental usage of it is a net-negative for them.
- vunderba 2y agoI don't know if that's true, necessarily, but I will note at least anecdotally I find that the larger my context window becomes the more often it seems to make mistakes. eventually, I just have to completely start a new chat even if it's the same topic and thread of conversation.