4 ms·

You see 20 of these posts a day and I have yet to see one that provides before and after prompts. Given ChatGPT saves your history, I don’t even understand why

• jstarfish 3y ago
  The problem can't be articulated in a single prompt response. There are no cases where it goes "full retard" in an obvious way that will convince skeptics "oh, i see." No cases of asking for a Python script and it returns a recipe for grilled snake. It's more like watching someone develop Alzheimer's or something over time. They just start forgetting small things and become harder and harder to work with. Which, to prove to you, one would have to show more of their history than they likely feel comfortable with. I'm the last person to excuse concealment of evidence but the nature of the proof you're asking for is invariably intrusive.
  • flangola7 3y ago
    It is intrusive but there's also no shortage of enthusiast users who doubtlessly have samples they would be comfortable sharing.
  • mnk47 3y ago
    But what if ChatGPT always had this issue, and it's now becoming more apparent because people are using it more? That sounds more likely to me. Maybe it's not like watching someone develop Alzheimer's, but more like watching a sloppy musician perform. At first it's impressive, but the more you listen the more you notice that he's constantly missing notes, and the longer he performs, more and more mistakes start pilling up.
  • cma 3y ago
    When they only confirmed the frozen models in the API hadn't changed and not the web interface I feel like that was pretty much admission by omission that chat gpt 4 under premium subscriptions had been lobotomized by sparsification or quantization or aomething.
• behnamoh 3y ago
  The onus is on ClosedAI to prove that chat.openai.com still uses the same gpt-4 as before.