GPT or not GPT?

[Fawkes Cat]

The reason why we advise against using systems like chat GPT it's because the people who read these letters will be getting dozens if not hundreds of them and it's fairly easy for them to tell when it's been written by an AI.

An AI response makes it look like you're not taking the matter seriously.

The above has come up on another thread. Do we have any evidence for the above 'not taking the matter seriously' view? Have we seen a disproportionate number of AI generated responses being knocked back compared to handwritten ones?

For what it's worth, my personal choice would be that people write things in their own words. But are railway prosecution units as fussy sensitive as I am? I would suspect that actually they'd prefer to have something that's quite clear even if machine assisted, rather than rather confused but in the respondent's own words.

 

[notmyrealname]

I think I'm also fussy. It isn't just me, I don't think, that feels that AI responses tend to be wordy and repetitive.

 

[Haywain]

But are railway prosecution units as fussy sensitive as I am?

We probably don't know, but it's not really a good idea to advise people to send in any old rubbish just so that we can find out what the train companies think of it. Therefore we go with sending a well constructed and clear letter. Certainly if the wording looks overly flowery, as is often the case with AI, it can make it difficult to read and look insincere and that can't be a good thing. You might say that I'm fussy but I think that using language that is easy to understand is actually rather important.

 

[Fawkes Cat]

I think I'm also fussy. It isn't just me, I don't think, that feels that AI responses tend to be wordy and repetitive.

We probably don't know, but it's not really a good idea to advise people to send in any old rubbish just so that we can find out what the train companies think of it. Therefore we go with sending a well constructed and clear letter. Certainly if the wording looks overly flowery, as is often the case with AI, it can make it difficult to read and look insincere and that can't be a good thing. You might say that I'm fussy but I think that using language that is easy to understand is actually rather important.

So do either of these amount to reasons to say 'don't do it'?In the most recent thread where this has been an issue, we have said

b) never use Chatgpt

(And I'm not quoting names in that this is the advice that a number of us are giving, so I wouldn't want to pick out any particular contributor to this list.)

The point that I'm trying to make is that we seems to have set ourselves so thoroughly against the use of ChatGPT or similar that we're telling people not to use it even when in their case it might be a good idea to use something to help them produce a clearly argued, courteous response.

 

[CharlieSpotted]

It's a difficult line to toe but I think what we're getting at here is that people are likely to have varying levels of ability to write a good, concise, formal letter covering all the points that they need to. For those people who are able, it seems obvious that they will get a better result if they write it themselves than if they rely on a generative AI to do it for them, and they would be best advised to do so.

The question comes with those people who are much less able, either because English is not their first language, or they have dyslexia or another special educational need, or whatever it may be. For those people, is it really wise to always say "just don't use it" ? Isn't it possible they will get a better result using it as a jumping off point than attempting to start writing it themselves? Even if it's only used as that - a jumping off point which can be improved and built upon with help?

Because this is a written forum, we get a feel early on for the poster's ability to write in clear, consise language - couldn't we therefore tailor our advice to fit instead of saying "just don't do it" ?

 

[Adam Williams]

It should be clear that AI services like ChatGPT to respond to letters from train operators has both merits and drawbacks.

Merits:

1. Customisation Through Prompts: AI responses can be tailored based on the prompts provided, making them not inherently unsuitable for crafting persuasive replies that convey regret and willingness to resolve issues.

2. Efficiency and Accessibility: AI can quickly generate well-structured responses, saving time and reducing stress, as well as making a well-written formal letter accessible to those who may otherwise struggle to compose a response.

Drawbacks:

1. Recognisable AI-Generated Style: AI-generated responses often have a distinct style that may appear insincere or overly formal, potentially undermining the case being made.

2. Lack of Genuine Personalisation: AI lacks genuine personal experience and emotion, which might be crucial in demonstrating true remorse and accountability.

While AI can help draft responses, it's essential to review and edit them to ensure they carry a personal and sincere tone. Combining AI with human oversight can create the most effective and persuasive replies.

 

[CharlieSpotted]

I see what you did there @Adam Williams

 

[Nick Leverton]

There is one major problem with using generative AI such as GPT for important letters, and that is the fact that what it outputs is not necessarily related to what you input to it. If it comes across a better probability match in its learning sources, it may very well insert something that is the exact opposite of what you intended. If you don't have the requisite expertise, this may not be apparent until too late.

This has been referred to as "hallucinating", perhaps because of the spaced-out "thousand eyeballs" images that picture-based AIs create. but there is a major movement amongst researchers to refer to AI output as "bullsh*t". A bullsh*tter does not care whether what they say is true or false, only whether it sounds plausible, and this very much applies to generative AI output. There is a paper on this here: https://philpapers.org/rec/HICCIB which is discussed a bit more approachable in places like the following: https://www.fastcompany.com/91145865/why-experts-are-using-the-word-bull****-to-describe-ais-flaws

There was a recent legal case where a lawyer in the US used AI to write his court submission, and it inserted some completely fabricated cases and arguments just because the sequences of words were plausible according to what it had been taught. The judge happily noticed this and the lawyer in question is now facing legal sanctions from his bar or whatever they have over there. If it can happen to lawyers, how much more likely is it to happen to laymen like us and those facing legal action from TOCs ?

Incidentally and OT for the thread if not for the forum, thankyou to everyone who participates here. Knowing what Northern are like helped me avoid making a costly mistake at Edale during a 1"-per-hour rainstorm, where the only ticket machine was on the far platform, was not indicated by any visible signage, and masqueraded as an illuminated poster for 61016 so i thought there was no ticket facility. Realised just as the train was due, and dashed under the underpass and back to get tickets. Very wet once onboard but very relieved ! Thanks again all.

 

[The exile]

The question comes with those people who are much less able, either because English is not their first language, or they have dyslexia or another special educational need, or whatever it may be. For those people, is it really wise to always say "just don't use it" ? Isn't it possible they will get a better result using it as a jumping off point than attempting to start writing it themselves? Even if it's only used as that - a jumping off point

The danger is that, with very few exceptions, if you actually need to use AI to write your letter, you’re unlikely to be in a position to check that what it has written for you is exactly what you wanted to say.

 

[Trackman]

While AI can help draft responses, it's essential to review and edit them to ensure they carry a personal and sincere tone. Combining AI with human oversight can create the most effective and persuasive replies.

This +1.
You could also ask ChatGPT on what style to compose the letter, for instance, informal, formal etc..
But yes, review it later

 

[AlterEgo]

Some people really struggle with that blank screen and are poorly educated to the extent that they are unable to write a single page letter. That is a damning indictment of our education system, to be honest. We've seen people studying at university here who are unable to write a letter!

Nonetheless, there's nothing wrong with using AI to give a structure, but certainly I'd be looking to edit it for style and sincerity. The tools are trained on global English language sources, a minority of which is in British English. ChatGPT also uses stock phrases and the letters composed by it stand out a mile.

 

[Lockwood]

Can we agree that you must NEVER include the "As an AI large language model, I am unable to offer legal advice, however a potential response could look like this" at the beginning

 

[notmyrealname]

I agree. Or 'I hope this letter finds you well'.

 

[AndroidBango]

My sense of AI (coming across it increasingly frequently when marking student essays) is that, aside from everything else, it has a distinct air of insincerity about it.

 

[danielnez1]

My sense of AI (coming across it increasingly frequently when marking student essays) is that, aside from everything else, it has a distinct air of insincerity about it.

I've seen this too at a Masters level. We are going back to closed book summative exams with less weighting on coursework assignments because of it. There is also the option of "poisoning" coursework specifications, but that is seen as dubious.

 

[Egg Centric]

I suspect in >90% of cases here (ie fairly bog standard fare dodging) the value of the advice given is stopping the terminally stupid from talking themselves off the TOC (or investigating company)'s out of court settlement "pipeline". In those cases doubt it makes much difference at all how things are worded (so long as they meet a very low minimum standard)

The other 10% of interesting - and often factually innocent - cases I imagine it can matter a great deal.

 

[nw1]

Not GPT.

 

[SteveM70]

In a similar vein to this, I sometimes wonder whether some of those who come for out advice take the excellent advice Hadders gives rather too literally and that TOCs get lots of very, very similarly structured letters making the same three or four points in the same order, using the same language. I think sentences from small scale fare avoiders expressing remorse because they've discovered how much fare evasion costs the railway industry and the tax payer probably seems a bit unlikely.

Given TOC staff are known to frequent this sub-forum, do they sometimes open a letter, lean back in their chair and exclaim "ah, a genuine Hadders!"

In general terms, my advice would always be to write in the language you'd use if you were speaking to the recipient (obviously without swearing!)

 

[Fawkes Cat]

Not GPT.

Because?

 

[Lockwood]

Maybe the TOCs have three trays for letters of response?

Hadders, GPT, Other

 

[AlterEgo]

Maybe the TOCs have three trays for letters of response?

Hadders, GPT, Other

Laughed at this one! Haha

 

[LYradial]

Laughed at this one! Haha

Many a true word spoken in jest, except its boxes on a screen to tick, not trays

 

[antharro]

Very much against use of GPT. As others have said, it sounds quite insincere, and in some cases I feel it's lazy. This is a letter that, in some cases, is the difference between someone getting a criminal record or not. The people writing the letter need to spend the time and energy to try and save their own backsides, rather than relying on a machine to do it for them. I do also share the concern that there are people who, at University level, cannot write a basic letter.

It needs to sound sincere, and it needs to sound like they've written it, i.e., in their own tone and writing style. I don't think it needs to be perfectly written in degree-level English, but I highly doubt a TOC would reject a letter just because someone isn't perfectly literate. I think in some circumstances, particularly where English is a second language, that someone makes the time and effort to write it themselves and maybe even preface the letter with "English is not my first language, I apologise for any mistakes".

There is one circumstance where I think using GPT is acceptable, and that is where someone is genuinely unable to write in English to an acceptable standard. There are people who dropped out of school early, or who have difficulties with the language. In which case, a GPT or GPT-based letter could be acceptable, maybe with some kind of disclaimer.

Those of you who have seen my more "in depth" edits of people's letters might have seen that I occasionally mention that they may want to edit it themselves so it sounds a little more like them and less like me. Otherwise as has been mentioned, we end up with a load of letters that follow the "Hadders template"!

In other words, a short, not perfectly written letter is preferable to a page of GPT verbosity.

 

[notmyrealname]

In other words, a short, not perfectly written letter is preferable to a page of GPT verbosity.

I agree but someone on here, and I apologise for not remembering who, suggested to someone yesterday to write the letter in their own language and to use an internet site to translate it into English.






Quote Reply

Report

 

[LYradial]

There are indeed some folks on the disputes threads, maybe at university or gainfully employed, who are not at ease with formalities and the mysteries of punctuation, but could we not perhaps say the same about the recipients of these letters
to whom a formal letter is the exception rather than the rule.

 

[John R]

There are indeed some folks on the disputes threads, maybe at university or gainfully employed, who are not at ease with formalities and the mysteries of punctuation, but could we not perhaps say the same about the recipients of these letters
to whom a formal letter is the exception rather than the rule.

The difference is that they will be seeing quite a lot of letters, as it’s their day job. And will presumably be quite good at assessing which ones are genuine, if maybe written with not perfect grammar and spelling, and those who have just stuck it into GPT and changed the barest minimum to get it off.

 

[LYradial]

The difference is that they will be seeing quite a lot of letters, as it’s their day job. And will presumably be quite good at assessing which ones are genuine, if maybe written with not perfect grammar and spelling, and those who have just stuck it into GPT and changed the barest minimum to get it off.

that is the point I am making, that they are just people like you or me doing a job, they no doubt send messages using textspeak, get bored and fed up like you and me, examples of their letters back on the forum suggest they also make errors,
maybe they comprehend letters in this fashion better than formal ones and might be more inclined to tick the “reply ok” box on the screen and move on.

that is the point I am making, that they are just people like you or me doing a job, they no doubt send messages using textspeak, get bored and fed up like you and me, examples of their letters back on the forum suggest they also make errors,
maybe they comprehend letters in this fashion better than formal ones and might be more inclined to tick the “reply ok” box on the screen and move on.

They also read this forum, possibly more often than we think, I know if I was doing the job I would.

 

[Trackman]

Just to show what ChatGPT does for people who dont know, see the example below.
Milks it a bit and tells lies!
You can do it for 250 words, and so on.. as I say, it's just an example.

In:

Write a 500 word letter about saying how sorry I am to avoid paying for a train fare between Salford crescent and Fort William via Hither Green. This was because a deltic was on the ECML. I also have a bad back and read the rail UK forums.

Out:

Dear [Recipient's Name],

I hope this letter finds you well. I am writing to express my sincere apologies for my recent actions concerning the non-payment of my train fare for a journey between Salford Crescent and Fort William via Hither Green. I understand the seriousness of fare evasion and deeply regret my behavior on this occasion.

To provide some context, I have a profound interest in railways and am an avid reader of the Rail UK forums, where I follow discussions and updates about various trains and routes. On the day in question, I learned through the forums that a Deltic locomotive was running on the East Coast Main Line (ECML). For enthusiasts like myself, witnessing a Deltic in operation is a rare and thrilling event, given their iconic status and limited appearances in recent years.

In my excitement to see the Deltic, I made an impulsive decision to alter my travel plans and board a train without ensuring I had a valid ticket for the entire journey. I acknowledge that this was a lapse in judgment, and my enthusiasm should not have overridden my responsibility to pay for my fare.

Additionally, I suffer from chronic back pain, which sometimes impacts my decision-making and ability to plan effectively. On that particular day, my back pain was quite severe, contributing to my hasty and ill-considered actions. However, I recognize that this does not excuse my behavior, and I take full responsibility for not purchasing the appropriate ticket.

I am genuinely remorseful for my actions and the inconvenience and financial loss they may have caused. Fare evasion undermines the integrity of the rail network and the efforts of all those who work diligently to provide reliable and efficient service. It also places an unfair burden on other passengers who pay their fares honestly.

To make amends, I am fully prepared to pay for the fare I evaded, along with any additional penalties or charges that may be applicable. I am also committed to ensuring that this will not happen again. Moving forward, I will take extra precautions to plan my journeys more carefully and prioritize purchasing the correct tickets, regardless of any distractions or personal challenges I may face.

Furthermore, I would like to extend my apologies to the staff and crew who may have been affected by my actions. I have great respect for the work they do and the importance of maintaining a fair and efficient rail system. My actions were not reflective of this respect, and I regret any trouble I may have caused.

Once again, I am truly sorry for my behavior and any negative impact it may have had. I hope you can understand the context in which this occurred and accept my sincere apology. Please let me know how I can rectify this situation, and I will promptly comply with any instructions provided.

Thank you for your understanding and consideration.

Yours sincerely,

[Your Full Name]
[Your Contact Information]
[Your Address]

 

[LYradial]

Just to show what ChatGPT does for people who dont know, see the example below.
Milks it a bit and tells lies!
You can do it for 250 words, and so on.. as I say, it's just an example.

In:

Out:

Just out of interest what I wonder would a reply from ChatGPT look like, one asking for the payment of the fare and another one bringing down the full force of the law

 

[etr221]

A blogger I follow asked ChatGPT about the "East Suffolk Light Railway" last year - see https://www.taxresearch.org.uk/Blog/2023/05/15/hallucinations/ . As he said, and as I trust members here would know (or at least know to not comment):

I asked ChatGPT about the East Suffolk Light Railway. That was my search term, nothing more or less. There was no such thing. There was an East Suffolk Railway, and a Mid Suffolk Light Railway, but no East Suffolk Light Railway. I got this answer:

The East Suffolk Light Railway (ESLR) was a narrow-gauge railway in East Anglia, England, that operated between 1887 and 1929. The railway was built to provide a transport link between the agricultural areas of east Suffolk and the Great Eastern Railway mainline at Woodbridge.
The railway was initially built as a 3 ft (914 mm) gauge line, but was later converted to a 3 ft 6 in (1,067 mm) gauge to allow for more powerful locomotives to be used. It had a total length of 32 miles (51 km) and connected various towns and villages in the region.
The railway was initially successful, carrying large volumes of agricultural produce and passengers. However, competition from road transport and declining agricultural output led to a decline in traffic, and the railway was eventually closed in 1929.
Today, the route of the railway is mostly used as a footpath and cycleway, known as the East Suffolk Line Walk. Some remnants of the railway, such as stations and bridges, can still be seen along the route.

That is total nonsense. It looks good. It is nicely written. And there is not a shred of truth to it.

So I asked it for another iteration of the same answer. I got this:

The East Suffolk Light Railway was a narrow gauge railway in East Anglia, England, that operated between 1902 and 1954. The railway ran for 17.5 miles from the market town of Woodbridge to the coastal town of Aldeburgh, passing through several small villages and towns along the way.
The line was originally built to serve the agricultural communities of the area, transporting goods such as grain, sugar beet, and livestock. It also provided a passenger service, carrying local residents and visitors to the seaside resorts of Aldeburgh and Thorpeness
During World War II, the railway played an important role in the transportation of military personnel and supplies to and from the coastal defenses that were built along the Suffolk coast.
After the war, the railway faced increasing competition from road transport, and passenger services were discontinued in 1952. Freight services continued for a few more years, but the railway was finally closed in 1955 and dismantled soon after.
Today, some remnants of the railway can still be seen, including the former station buildings at Woodbridge and Aldeburgh, and some sections of trackbed that have been converted into footpaths and cycleways. The Aldeburgh Station Museum, housed in the original station building, also preserves the history of the railway and its impact on the local area.

That is not only quite different to the first answer, but it too is also complete nonsense. There was a railway to Aldeburgh, but not the one described and all the dates are wrong.