4 ms·

I salute you for being open about the chatgpt use! Do you really trust what it says? That quoted chatgpt-number has zero value to me. The other numbers are val

• pqtyw 1y ago
  What if the links came from the ChatGPT answer as well?
  • mpweiher 1y ago
    They did.
  • Epa095 1y ago
    For me that does not change anything really. I assert the trustworthyness of the webpage as usual. My problem is the statement from chatgpt. I have seen it invent enough bullshit that if it was a person I would have labeled them as untrustworthy a long time ago. Yes yes, it's also amazing and all that jazz, but I still don't know how to trust a 'Chatgpt told me this' - quote.
    • pqtyw 1y ago
      I do get it. However it's hardly any different or less trustworthy than a random person making a random claim identical to what ChatGPT would say. Of course a 'Chatgpt told me this' disclaimer does indicate something, i.e. either that person has no clue about the topic and is unable to verify the answer at least to some extent on their own and is just blindly copy pasting something and/or believes that anything LLMs say is inherently credible on its own without extra verification.
      • mpweiher 1y ago
        Good grief. It's exactly the opposite.
  • Gud 1y ago
    Then why even bother with the "I asked chatgpt"? Just cross reference the links and credit the original sources. It just adds verbosity and doubt to the statement.
  • subscribed 1y ago
    It's well known for making the stuff up because this is how it works The Guardian found a article attributed to them, generated (not "written") by chatgpt. A silly lawyer got into trouble trying to use chatgpt-generated precedences in court. Everything chatgpt prints out is made up, and that includes links. Seriously, heed the warning the company itself prominently prints I app and in the webui. Chatgpt may print out mostly true made-up sentences, but by definition, because oh how it works, it doesn't generate truth. It generates tokens that make up words. Chatgpt is not a RAG, come on, it's 2025!
    • pqtyw 1y ago
      It can (and does actually) open and verify the links it provides so it's not as bad anymore, at least when it's using real existing articles/papers/etc. it find as sources inside its context.
• mpweiher 1y ago
  > Do you really trust what it says? "Really trust"? Nope. But I think it gives me a good ballpark estimate and ways to check if that estimate is about right or not. Checking the answer is quicker and potentially less error-prone than compiling the answer.
  • Epa095 1y ago
    Ok, so did you verify the 500 billion? If so, then that's really the relevant part for me. But then I trust you, not Chatgpt.
    • rad_gruchalski 1y ago
      Where did you take the 500 billion figure from?
      • Epa095 1y ago
        It's from the chatgpt quote my comment was an comment to: "[...] in 2023 was approximately €505 billion"
        • rad_gruchalski 1y ago
          Thanks.
  • subscribed 1y ago
    Unless you verify the ballpark figure you shouldn't really use it in the conversation. Chatgpt is a glorified autocomplete. Don't share it's output unverified like that was some sort of an oracle. If you really HAVE TO resort to "AI", at least use Perplexity.