Yes ChatGPT’s ability to pull a quote from a library of information based on a vague concept is like using Google for the first time and realizing it’s capability.
1 Like
I also cannot imagine what librarian would not appreciate the ability to conceptually search the text of an entire library. ChatGPT’s faults and obvious lapses in judgement will continue to be corrected into the foreseeable future
1 Like
One thing to remember about it and current events is that it does not have access to recent stuff. I think it stopped back in 2020 or maybe 2019 or something. I don’t remember.
3 Likes
Yes, it went into an explanation of how it works.
Okay, that’s not actually the same thing that I asked. There’s a difference between constructing a mathematical infinite series and making an infinite number of additions in the real world.
2 Likes
Only a human could mess that up 
In many ways, ChatGPT seems to be designed with the Turing test in mind. But we already have plenty of humans; we do not particularly need human-like computers. Doing tasks well that humans can’t do would be useful. Imitating a human well is what makes Chat GPT function so much as CheatGPT. Although it can do some things well, error checking is not one of them.
4 Likes
I wonder if we become fascinated by this program and more inclined to view its output as profound to the degree we harbor a scientistic POV, in other words the view that the world is mechanistic and that truth is best arrived at by computation.
1 Like
I’m of the mind that ChatGPT is a tool. And like any new tool, the skill is in knowing when to use it and when not to. I probably use CHatGPT a dozen times a day for everything from language practice, TTRPG content creation, learning, a fast ‘Google’, work tasks, copy editing, etc. I resonate with its language style, and as a person who learns through discussion I find it super helpful. Yet, as in the old saying about everything looking like a nail when all you have is a hammer, problems arise when it is viewed as the tool to end all tools.
There is nothing wrong with using ChatGPT to research topics, and allow that content to inform posts. We all do that I am sure with books and the internet, which are just as fallible as current generative AIs. However, one wouldn’t respond to a post by copying/pasting a paragraph from a book or Wikipedia, with little to no context or caveat. Which is the problem we are addressing here.
That said, despite some obvious concerns, I am overall incredibly thankful for the work of OpenAI in providing this amazing tool and am excited about what the future holds.
7 Likes
Exactly. You also don’t just throw them out because you don’t know how to use them.
That’s not what I see being addressed in the new directive.
And I don’t think it’s as big of a problem as it’s made out to be. Context is king. I don’t recall that many posts which contained a ChatGPT quote minus commentary. There may have been a couple thrown out there by myself without commentary but these should not have been addressed to a forum user in particular.
Additionally, even when quoting from Wikipedia, one has references for the source of information. People can follow up and examine the quoted pages (or articles, or database or whatever) and the sources used to create that page. ChatGPT does not (automatically) provide a list of references, and does not always give the same answer, which indicates it probably doesn’t always use the same sources.
2 Likes
It is cool that it will provide them quickly upon request though. Since it’s a conversational model it mimics us – we don’t have superscript smaller font numbers in little speech balloons referring to footnotes while we are conversing. 
It certainly does not learn from or refer to prior conversations anyway.
And it will readily admit when it is wrong if you can demonstrate it. That is not common in live conversations!
1 Like
Good point. It would be helpful if there was a way to provide a reference for a ChatGPT generated response when there is a real insight or something particularly interesting. Being able to save and source the quote would be a nice addition to the platform.
I think, based on my naive understanding of machine learning, it has access to the same library and sources in any given instant. I wonder what accounts for the variation of responses. It’s not like it is using a pseudo-random algorithm to vary it’s pull on data.
1 Like
ChatGPT confirmed that it does treat it’s sources as a single source. It explains, “the model is trained on the entire corpus of text as a single entity, without differentiating between the individual sources of text that make up the corpus.” It goes on to clarify how some variations and patterns can affect responses.
Simply amazing that an entire library can now be accessed with such rudimentary ease.
2 Likes
It is indeed what is being addressed in the new directive. We don’t want any more posts of the nature "I asked Chat GPT this question and this is what it said, “XYZ”
2 Likes
Even when we’re talking about its usefulness as a tool, illustrating its features and quirks, in a thread about ChatGPT.
Take down the BioLogos interview with it then.