| Σελίδες για το θέμα: < [1 2 3] > | People answering questions with AI output Αποστολέας σε συζήτηση: Zea_Mays
|
|---|
In Portuguese Ai is an interjection that expresses pain or joy…
| | | | Tom in London
Ηνωμένο Βασίλειο
Local time: 11:13
Μέλος από 2008
Ιταλικά σε Αγγλικά
| that post was clearly written by AI | Jul 12, 2023 |
Zea_Mays (or rather AI) wrote:
all that stuff in the previous post
which was clearly written by AI
| | | | Philip Lees 
Ελλάδα
Local time: 13:13
Ελληνικά σε Αγγλικά
Maria Teresa Borges de Almeida wrote:
In Portuguese Ai is an interjection that expresses pain or joy…
And "ai" is another name for the pale-throated sloth.
Isn't that interesting?
| | | | Lieven Malaise
Βέλγιο
Local time: 12:13
Μέλος από 2020
Γαλλικά σε Ολλανδικά
+ ...
Zea_Mays wrote:
It's worth noting that AI models like ChatGPT are continually improving, and there are ongoing efforts to make generated text more indistinguishable from human-written text. As a result, these indicators may become less reliable over time.
I am currently translating software content for a company specializing in plagiarism and AI writing detection. It's an interesting battle.
| | |
|
|
|
Zea_Mays
Ιταλία
Local time: 12:13
Αγγλικά σε Γερμανικά
+ ...
ΞΕΚΙΝΗΣΕ ΤΟ ΘΕΜΑ
Tom in London wrote:
Zea_Mays (or rather AI) wrote:
all that stuff in the previous post
which was clearly written by AI
Didn't you say no one can say if a text was written by AI?
| | | | Elena Feriani
Ιταλία
Local time: 12:13
Μέλος
Γαλλικά σε Ιταλικά
+ ...
| 40+ language pairs | Jul 13, 2023 |
I saw someone answering questions in more than 40 language pairs, including EN-IT... Some of their answers looked pretty good to me... I thought they just had a lot of friends
| | | | | Human judgement vs. algorithm | Jul 13, 2023 |
Zea_Mays wrote:
Tom in London wrote:
Zea_Mays (or rather AI) wrote:
all that stuff in the previous post
which was clearly written by AI
Didn't you say no one can say if a text was written by AI?
I don't believe anyone said that a human can't form a judgement as to the origin of a text. What's being said, or at least implied, is that no one at ProZ has the time to read every answer and decide whether it's AI or not, so it would have to be done by an algorithm. And posters are stating that it is extremely difficult to write an algorithm that can differentiate between an AI-generated and a human-generated text.
| | | | Philippe Locquet
Πορτογαλία
Local time: 11:13
Μέλος από 2013
Αγγλικά σε Γαλλικά
+ ...
| Context, for asker and for AI | Jul 13, 2023 |
Zea_Mays wrote:
- Lack of contextual awareness: While AI models excel at generating text, they might sometimes fail to fully understand the context or nuances of a conversation. This can lead to generic or off-topic responses.
No. As I demonstrate in a recent video with Bing (free) https://youtu.be/u4nROnmnIxI , you can feed LLMs context very efficiently, this is a main difference with MT (needs to be retrained, or other systems to take context into account).
So it wouldn't be hard with the context provided by the asker to create a prompt asking an LLM to create an output according to that specific context.
- Overuse of certain phrases: AI models might exhibit repetitive patterns, use specific phrases, or rely on templates more frequently than a human would.
Yes, that may be used to our advantage with Kudoz.
- Output length and speed: AI models can generate lengthy and detailed responses quickly, which may be an indication that the text was generated by an AI rather than a human who typically takes more time to compose a similar response.
True!
Bearing in mind the last two points, if an asker wants human answers only, they could ask first theLMM for an output (maybe it will give them what they're looking for, which would save time) and then paste the LLM's output in their comments when posting. It would be quite fast to do and as robots have a tendency to repeat the same thing (or a close variation) it would be easy to weed out non-human answers in the event where the LLM wouldn't be right (if it gives the correct answer, then it's likely that human answers will be very close or identical to the LLM).
Challenging!
| | |
|
|
|
| Context and AI | Jul 13, 2023 |
Philippe Locquet wrote:
Zea_Mays wrote:
- Lack of contextual awareness: While AI models excel at generating text, they might sometimes fail to fully understand the context or nuances of a conversation. This can lead to generic or off-topic responses.
No. As I demonstrate in a recent video with Bing (free) https://youtu.be/u4nROnmnIxI , you can feed LLMs context very efficiently, this is a main difference with MT (needs to be retrained, or other systems to take context into account).
So it wouldn't be hard with the context provided by the asker to create a prompt asking an LLM to create an output according to that specific context.
As AI does not have a conscious mind, it still doesn't manage to make correct use of such context all the time. Here is a practical example with ChatGPT from yesterday:
I asked it (the whole conversation was in Danish): 'Are there any Danish rules for how footnote cues and the footnotes themselves are written?'
It listed some rules and then proceeded to say: 'It's important to note that there may be variations in practice depending on the writing style you follow, such as Chicago Manual of Style, APA or MLA.
Style guides for English obviously have no relevance for Danish, which it admitted when I pointed it out.
Then I asked it for the source of the rules it had listed.
'The main source of the Danish spelling rules is "Dansk Retskrivningsordbog", published by Dansk Sprognævn,' it said. That is indeed the official Danish orthography and style guide. I had just been unable to find anything about footnotes in it.
As it was sidestepping the question, I asked it again which chapter in that book contained the rules it had just quoted.
Reply: 'I apologise for the confusion in my previous answer. As an AI model, I do not have direct access to specific sources such as the "Dansk Retskrivningsordbog" or the ability to quote specific paragraphs from it. I cannot provide specific references or paragraphs from the source.
My answers were based on general guidelines for footnote referencing and footnote formatting that are widely recognised in academic writing across different language areas. I apologise for giving the impression that these guidelines are specifically laid out in the "Dansk Retskrivningsordbog".'
Even a teenager would instantly have understood that quoted American style guides for English when the question is about Danish is completely irrelevant, but AI still cannot handle such elementary logic.
Another problem is the difficulty of getting a precise answer with source to a precise question. AI tends to produce a lot of waffle that doesn't answer the direct question.
This is why posting AI answers in a language one does not understand could easily make that person look like a clown (or a politician, as politicians in particular are adept at sidestepping questions).
Bearing in mind the last two points, if an asker wants human answers only, they could ask first theLMM for an output (maybe it will give them what they're looking for, which would save time) and then paste the LLM's output in their comments when posting. It would be quite fast to do and as robots have a tendency to repeat the same thing (or a close variation) it would be easy to weed out non-human answers in the event where the LLM wouldn't be right (if it gives the correct answer, then it's likely that human answers will be very close or identical to the LLM).
Who or what would 'weed out' such answers with which resources? And is this so important that it could justify significant and costly software development efforts or the working time of moderators (who would not always get it right)?
| | | | expressisverbis
Πορτογαλία
Local time: 11:13
Μέλος από 2015
Αγγλικά σε Πορτογαλικά
+ ...
Elena Feriani wrote:
I thought they just had a lot of friends
No, they have just a lot of nerve!
It must be that same member with more than 5000 language pairs...
On Kudoz, one thing is to get answers from colleagues who are language experts and master a language inside and out, regardless of whether they use (wisely) AI or other resources (after all, AI can be seen as a tool); another thing is to receive answers from someone who does not have the slightest knowledge of certain languages and "abuses" AI and other tools. IMHO, this is almost disrespectful.
| | | | Zea_Mays
Ιταλία
Local time: 12:13
Αγγλικά σε Γερμανικά
+ ...
ΞΕΚΙΝΗΣΕ ΤΟ ΘΕΜΑ
if it would make a difference or not if I make a suggestion stating that I used AI to produce the answer.
An idea for when Proz will be able to use an AI content detection tool: labelling such answers with a warning like "This content may have been generated by an AI bot".
| | | | Zea_Mays
Ιταλία
Local time: 12:13
Αγγλικά σε Γερμανικά
+ ...
ΞΕΚΙΝΗΣΕ ΤΟ ΘΕΜΑ | It's like in any sport or game | Jul 13, 2023 |
Thomas T. Frost wrote:
Who or what would 'weed out' such answers with which resources? And is this so important that it could justify significant and costly software development efforts or the working time of moderators (who would not always get it right)?
There are rules and, most of all, an explicit or implicit commitment to honesty and fair play.
Btw: Posting AI generated answers is done to collect points, and points as far as I know help you gain more visibility over time.
| | |
|
|
|
| In Italian too | Jul 17, 2023 |
Maria Teresa Borges de Almeida wrote:
In Portuguese Ai is an interjection that expresses pain or joy…
Ai que dolor! (che dolore)
| | | | Schtroumpf
Local time: 12:13
Γερμανικά σε Γαλλικά
+ ...
| Ban AI trolls, please! | Jul 17, 2023 |
Thanks ZeaMays for raising the topic. I strongly support the idea of banning contributors who post (mainly or exclusively) AI generated stuff.
There is no need in my opinion that Proz detect these posts automatically. If we happen to see any such content by let's say, some lady from far abroad indicating that one of her mother tongues is English (French variant - no joke) and answering in more than 40 language pairs, shouldn't we just report the case to staff and get the account deleted?... See more Thanks ZeaMays for raising the topic. I strongly support the idea of banning contributors who post (mainly or exclusively) AI generated stuff.
There is no need in my opinion that Proz detect these posts automatically. If we happen to see any such content by let's say, some lady from far abroad indicating that one of her mother tongues is English (French variant - no joke) and answering in more than 40 language pairs, shouldn't we just report the case to staff and get the account deleted?
There might even be no need to change any Proz rules at all. As someone pointed out lately, Proz translator accounts are intended for individuals and not for translating plants with 150 chimneys and 700 work stations, all of which run by people who never acquired any of the dozens of languages they pretend to know. ▲ Collapse
| | | | expressisverbis
Πορτογαλία
Local time: 11:13
Μέλος από 2015
Αγγλικά σε Πορτογαλικά
+ ...
| Not quite true | Jul 17, 2023 |
Thomas T. Frost wrote:
Askers can restrict their questions to natives or users who work in certain languages pairs. If they don't do that, they may get all sorts of unwanted entries.
That is not quite true. I have posted a question today, and restricted it to natives only. I got an answer from that same member with tonnes of languages... she is not a PT-PT native.
Despite everything, I still managed to smile when I saw the answer because she has no idea what false friends are in Portuguese and French.
AI is useful when it falls into the right hands, if it falls into the hands of opportunists the result is this one.
I can use it for doubts and questions of my own and if I really need it (for example, in case clients can't provide me with an answer), but I would never do it to try to answer colleagues with the sole purpose of collecting points or for whatever reason.
[Edited at 2023-07-17 16:56 GMT]
[Edited at 2023-07-17 17:51 GMT]
| | | | | Σελίδες για το θέμα: < [1 2 3] > | To report site rules violations or get help, contact a site moderator: You can also contact site staff by submitting a support request » People answering questions with AI output | Trados Studio 2022 Freelance | The leading translation software used by over 270,000 translators.
Designed with your feedback in mind, Trados Studio 2022 delivers an unrivalled, powerful desktop
and cloud solution, empowering you to work in the most efficient and cost-effective way.
More info » |
| | Wordfast Pro | Translation Memory Software for Any Platform
Exclusive discount for ProZ.com users!
Save over 13% when purchasing Wordfast Pro through ProZ.com. Wordfast is the world's #1 provider of platform-independent Translation Memory software. Consistently ranked the most user-friendly and highest value
Buy now! » |
|
| | | | X Sign in to your ProZ.com account... | | | | | |
Login to reply/comment 
