#References in Chat AI responses
19 messages · Page 1 of 1 (latest)
I asked it to generate an article with in text references and a bibliography in APA style. 2 out of the 3 references I got were existing (academic papers or scientific books). One was made up but convincing. I could not check if the sentences were indeed belonging to the in-text references either because I did not want to read a 200p book or I did not have access to the full article .
Interesting -- in my opinion, it's a problem if it is making up references, even worse if they are convincing rather than obvious placeholders.
In some respects, also worse if some articles are authentic and others contrived. It creates a problem of knowing what is accurate and confuses credibility.
I have the same problem. i asked to list some references about a topic and the AI return fake bibliography
i verifyed 3 of 4 articles and none are real.
the problem is it dosnt actually have access to any specific resources, it just has the raw info
Isn't it possible that the raw info it has includes academic papers? In which case I can understand how it came up with a list of references, although it might not know where these papers cam from so it might not be able to provide links to the papers. That said, its very concerning if its making up some of these references. FYI, after having a discussion with ChatGPT regarding whether or not gravity's influence has limits, I asked if it could site the references it used to answer my questions, but it said it was not able to (see attached).
but the AI has a database (a memory) of what was processed (and classified). So, it must return valid references or say that don´t know (or remember) the references about the subject.
🥅 Goal! Found something that i can change, that was bothering me
what is happening im not able to chat this morning with gpt
Do you know something about gender equality policies and their application in El Salvador?
how to call a vf from lwc
I'm a doctoral research student, I have the same issue, my guess is that the references are generated randomly based on topic, though if there is a DOI it is 50/50 an active link to a differnt article or nonexistant. I think it is something in the algorithm that creates a "like" source without validating it agianst actual documents.
I started using this prmpt for google scholar and pubmed...
"create a list of 5 journal articles on the topic of "Artificial intelligence ethical concerns," validate the references against Google Scholar for accuracy, APA format, and verify a link to the article from google."
this seems to work, I will continue to play with the prompt string. Would love to hear how others are using this...
done