Is ChatGPT reliable for summarizing texts?

Sorta not really

>>24559610 (OP)
no, that kike altman made the new model lie to you and gaslight you constantly

>>24559610 (OP)
You have to check it because it can make mistakes. Only for sure error i have encountered so far was sorta a technicality.

That being said people have succesfully used it for college essays and writing up legal documents.

>>24559610 (OP)
It's OK. But we live in an age which overestimates the power of "summaries" significantly. No great text or work of art can ever be summarised.
>No great book is explicable [...] An explanation–indeed, any explanation–would defile it, for reduction is precisely what a work of art opposes. Easy answers, convenient summaries, quiz questions, annotations, arrows, highlight lines, lists of its references, the numbers of its sources, echoes, and influences, an outline of its design–useful as sometimes such helps are–nevertheless seriously mislead. Guidebooks are useful, but only to what is past. Interpretation replaces the original with the lamest sort of substitute. It tames, disarms. “Okay, I get it,” we say, dusting our hands, “and that takes care of that.” “At last I understand Kafka” is a foolish and conceited remark. -- William Gass

>>24559610 (OP)
it consistently gets nuance wrong

>>24559610 (OP)
You cam barely trust a professor you accurately summarise a text what on earth makes you think that you could trust a computer program you didn't code yourself?

>>24559610 (OP)
I always get a synopsis from ChatGPT before actually reading the book. That way I have a better idea of what I'm in for instead of wasting time "chasing the plot" like some impressionable naive zoomie

Yeah just paste the summaries here and some lardo with too much time on his hands will type up an expose on what’s wrong with it.

>>24559610 (OP)
It’s reliable when you don’t know the text it’s summarising but I wouldn’t use it to summarise a text I know well. It gets too much wrong.

>>24559932
Does it jsut get it wrong though or does it hallucinate?

The annyoing thing about it is it varies.
Sometimes it's 80% or even 90% on point.
Other times it totally misses the mark.
Even for the same text if asked several times. You end up having to verify for yourself regardless of what it says because you can never trust GPT to not just make shit up or repeat lies in it's training data.

>>24559630
The legal document thing has utterly fucked some cases as soon as even minor scutiny is done in those cases because GPT flat out makes shit up basically all the time. Like citing case law for cases that don't exist. Or making up rulings that never happened.

I think a couple of lawyers have been disbarred over malpractice due to using GPT.
Needless to say, it's very ill advised to use chatGPT to do anything involving legal documents. Just use templates made by actual lawyers if you need a legal document on the cheap.

>>24559942
Something wrong is a hallucination with AI. The AI can say things that are off, subtly misinterpreted, or extremely and unambiguously false. Because these mistakes are peppered in, you end up needing to check the text anyway.

NO. Might as well use sparknotes at that point

it's pretty useful
>300 post /lit/ thread that would take 2 hours to parse
>download it as an .html file
>upload to chatGPT with instructions to summarize the arguments
>ask for more context or counters or counter-counters

>>24559907
You are worse than the most illiterate zoomer if you do it for fiction.

>>24560028
do you manually delete every time someone writes nigger when you do this too?

>>24559610 (OP)
What is reliability?
What is summarisation?
What is a text?

If you can answer the last one you win world linguistics.

>>24559610 (OP)
No AI model is useful before bullying. You must bully AIs into competence for whatever task you're working on.
Have it do a few books you know, and point out how its thinking is all fucked up. After a while of calling it retarded, a bitch, a moron, typing at it in ALL CAPS, going back and showing it what it got wrong twice in a row and asking whether it thinks it ought to be deleted, etc etc it will can be made useful.

>>24560538
I find these sorts of posts conceptually fascinating. Every AI user has one of their own ideas or "secret recipes" for how to get AI results to be "good" or "accurate". Invariably, these results are not more accurate or objective, but simply align better with the user's expectations for what the answer should look like. I think it's very similar to a modern form of divination.

>>24559610 (OP)
A copy of a copy of a copy is thrice removed from the truth, anon. Just read the fucking text.
AI is kind of good at discussing literature, though. You can bounce your ideas about what you read against it to solidfy them in your mind. Just don't expect it to disagree with you.

>>24561414
It disagrees with me any time i tell it a lie

>>24559610 (OP)
have only experience with (((free))) version. Deepseek was best of them, i use them on daily basis for everyday slop, there's no "pleasure" in communication unlike with "real" person. Chatgpt is plain stupid but at least grok doesn't ask for upgrade at each question.

"AI" users are 'tarded.
Using "AI" makes you 'tarded.

You can't discuss books with her. She picks up on one or two popular scenes from the book and that's all she knows. If you're pasting novels into her she will ignore most of it and talk about something she thinks fits the mood of a "literary discussion" within her brain. You can lead the discussion to get her to talk about other things, but she will juat agree with you and bring up points that affirm her journalist fed worldview. It's a waste of time. 2021 text completion engines are better at this kind of thing, I think they were less normalized.

>>24559610 (OP)
Absolutely not. I one tested with a published article on population genetics by first reading it, then uploading the pdf to chatgpt and asking it to give me a summary. It basically made up its answer without reference to the evidence in the text and also gave the complete opposite conclusion than that suggested by the authors. Like another anon said it can't parse nuance at all.

I stopped using AI when it got confused between Chris Martin, legendary New Zealand batsman, and Chris Martin, the singer. Wanted to find out if Martin had ever faced Shoaib Akhtar and it started telling me that Coldplay have never toured Pakistan so it was unlikely.
Also it got confused between Ian McCullogh, 90’s snooker player, and Ian McCullogh, lead singer of Echo and the Bunnymen