Re: [JVF] Abridged summary of jules-verne-forum@googlegroups.com - 2 updates in 1 topic

33 views
Skip to first unread message

quentin skrabec

unread,
Feb 10, 2025, 6:59:01 PM2/10/25
to jules-ve...@googlegroups.com
Truly hate the idea of AI translations. As a researcher I ve come to see the very unique skills and knowledge in translators --  it way way too early in AI to even think about Verne translations- i would never trust AI translation in my Verne technology research - i could see it creating more problems
Quent 


From: jules-ve...@googlegroups.com <jules-ve...@googlegroups.com>
Sent: Monday, February 10, 2025 6:43 PM
To: Abridged recipients <jules-ve...@googlegroups.com>
Subject: [JVF] Abridged summary of jules-ve...@googlegroups.com - 2 updates in 1 topic
 
Alex Kirstukas <alex.ki...@gmail.com>: Feb 10 10:43AM

I’m afraid I have to echo Tad’s skepticism. With the rise of generative AI, there’s been a huge influx of people trying to make a fast buck by feeding a public-domain text into ChatGPT or ...more
Tad Davis <tad.dav...@gmail.com>: Feb 10 06:42PM -0500

Not to beat a dead horse, but here's an example of the kind of incoherence that results when you publish so many translations so quickly, however they're generated.
 
In chapter 2 of Twenty Thousand ...more
[./#digest_top]Back to top
You received this digest because you're subscribed to updates for this group. You can change your settings on the group membership page.
To unsubscribe from this group and stop receiving emails from it send an email to jules-verne-fo...@googlegroups.com.

Rafael Ontivero

unread,
Feb 11, 2025, 5:14:30 AM2/11/25
to jules-ve...@googlegroups.com
Related to AI, I'm doing an unsuccessful experiment. I run in Claude 3.5 a very specific prompt to get all the toponims where the characters pass by, with the phrase in what it is named, from the French version of the novel. I got an extensive list of places... After that, I started reading the book (in Spanish, I don't read French), and in the 3 first pages, it forgot at least 3 places.

Uploading a novel (in French) and asking stuff from it, answers never are complete. As a sample, in Five Weeks, I asked how many females appear in the book, and it answer African women in battle scenes and a native character's wife (that I don't remember if it appears in the book), and forgot Dick Kennedy maid, and when asked about her, answer was something like "ah, sorry, I forgot, yes she is blah blah blah"...

Today AI is not ready for real academic work. One friend of mine, biology investigator, cries of despair when he tries to get right citations from AI.




De: jules-ve...@googlegroups.com <jules-ve...@googlegroups.com> en nombre de quentin skrabec <qrsk...@gmail.com>
Enviado: martes, 11 de febrero de 2025 0:58
Para: jules-ve...@googlegroups.com <jules-ve...@googlegroups.com>
Asunto: Re: [JVF] Abridged summary of jules-ve...@googlegroups.com - 2 updates in 1 topic
 
--
You received this message because you are subscribed to the Google Groups "Jules Verne Forum" group.
To unsubscribe from this group and stop receiving emails from it, send an email to jules-verne-fo...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/jules-verne-forum/DM6PR02MB56423B6367CF9B54AC2D09C4FFF22%40DM6PR02MB5642.namprd02.prod.outlook.com.

James D. Keeline

unread,
Feb 11, 2025, 10:33:23 AM2/11/25
to jules-ve...@googlegroups.com
From results I have seen, most of the generative AI systems are Large Language Models (LLMs) which are OK for paraphrasing existing content but take the the philosophy my wife likes to quote:

"English major, you do the math."

Counting and arithmetic and numerical comparisons are weak parts of AI systems.  Related to this, they are not great with logic either.  On the illustration systems, you start getting impossible numbers of fingers or typewriter keys or physically impossible or impractical things in images.

But lazy people love the idea of some computer doing the work for them.  They don't realize, what wiser people do, is that you have to run a prompt several times to get a passable result.  The propensity to make things up (euphemistically called a hallucination) and having to check every detail of the work shows where the systems are not all that people hope they will be.  I don't know if they will or can improve in these areas.  Right now they seem to be throwing more texts at them to train them and that means the garbage in garbage out principle applies.

It comes down to whether you'd use your math-challenged 10-year-old to do your income taxes.

If these mass translations are on Amazon, the thing to do is to review them and report that they are bad AI translations.  I would note that Amazon's Kindle Direct Publishing (print on demand division) has rules requiring that an "author" inform them of AI content so they can tag it and make decisions about it.


Artificial intelligence (AI) content (text, images, or translations)

We require you to inform us of AI-generated content (text, images, or translations) when you publish a new book or make edits to and republish an existing book through KDP. AI-generated images include cover and interior images and artwork. You are not required to disclose AI-assisted content. We distinguish between AI-generated and AI-assisted content as follows:

  • AI-generated: We define AI-generated content as text, images, or translations created by an AI-based tool. If you used an AI-based tool to create the actual content (whether text, images, or translations), it is considered "AI-generated," even if you applied substantial edits afterwards.
  • AI-assisted: If you created the content yourself, and used AI-based tools to edit, refine, error-check, or otherwise improve that content (whether text or images), then it is considered "AI-assisted" and not “AI-generated.” Similarly, if you used an AI-based tool to brainstorm and generate ideas, but ultimately created the text or images yourself, this is also considered "AI-assisted" and not “AI-generated.” It is not necessary to inform us of the use of such tools or processes.

You are responsible for verifying that all AI-generated and/or AI-assisted content adheres to all content guidelines, including by complying with all applicable intellectual property rights.

I don't know if Amazon presents the AI tags to shoppers.  But if they are using AI for translation, and have not disclosed it, they may be in violation of the KDP terms of service cited above.  There are a lot of qualifications there.

James D. Keeline



Tad Davis

unread,
Feb 11, 2025, 12:36:31 PM2/11/25
to jules-ve...@googlegroups.com
I suspect, but I don’t have to expertise to prove this is AI-generated text. I wouldn’t want to get into a legal battle over it. I think it would require someone to identify which engine was used and generate nearly identical results for selected texts; or produce output from one of the standard plagiarism tools (which have problems of their own, as I understand it). 
 
Tad

— 
Tad Davis
tad.dav...@gmail.com

Garmt de Vries-Uiterweerd

unread,
Feb 11, 2025, 12:42:54 PM2/11/25
to jules-ve...@googlegroups.com
For an article in the Dutch magazine De Verniaan, I experimented with AI translations. I fed various fragments from JV novels to Google Translate, DeepL and ChatGPT, letting them translate from French to Dutch, and compared results between the AIs and also to existing translations.

At times, one or two of the AIs (but not always the same ones) did better than the professional translators, but more often they messed up, misunderstanding words (both technical terms and everyday expressions), stumbling over French homonyms, missing context, or simply producing sentences that didn't parse at all.

A complicating factor when translating between two languages is that some AIs use English as an intermediate stage. If you ask Google Translate to translate "La voisine est gentille" to Dutch, it comes up with "De buurman is aardig". Both French "voisine" (feminine) and Dutch "buurman" (masculine) are gender specific, but because the translation goes via the English gender-neutral "neighbour", the machine translation loses this information.

I'll run the same tests for FR-EN translations and post the results here.

Cheers,
Garmt

Garmt de Vries-Uiterweerd

unread,
Feb 11, 2025, 12:47:46 PM2/11/25
to jules-ve...@googlegroups.com
I have the same issues with students who hand in AI-generated text for exam papers. I immediately recognize the typical ChatGPT style, but it can be hard to prove. Unfortunately for the students, I have a few tricks up my sleeve when it comes to physics reports with formulas etc., and I am smarter than they are in using Google Docs' version history :)



Reply all
Reply to author
Forward
0 new messages