Skip to main content

Hello everyone I spent some time the other day running a series of tests on the new AI chatGPT site that was launched in late November 2022 that is supposed to be a Google competitor.   

I initially ran some tough engineering related questions and then turned its attention to trains.

You'll note I asked the age old question about the most powerful and largest steam locomotive knowing it would be a tough question.  And for fun I asked it to generate a couple poems.   

Below are some questions I posed to it along with its answers.

In my opinion, this tool has a long way to go as its responses were in many cases incorrect.    Thus don't put much faith into what it spits out unless you already know the answer.

I hope you enjoy the questions and the OGR poem.

My questions are noted by the red box and the tools responses in green.

Note that all responses by the tool were relatively immediate to a couple seconds after I struck the enter key.

    You will note that the same question was posed again as the tool simply stopped working when I posed a new one.  This happened on several occasions when the questions were a bit tough.  When I asked about its sources, it faltered.

                                                                                                                                                                                                        

Attachments

Images (18)
  • mceclip0
  • mceclip1
  • mceclip2
  • mceclip3
  • mceclip4
  • mceclip5
  • mceclip7
  • mceclip8
  • mceclip9
  • mceclip10
  • mceclip11
  • mceclip12
  • mceclip13
  • mceclip14
  • mceclip15
  • mceclip16
  • mceclip17
  • mceclip18
Original Post

Replies sorted oldest to newest

This is really interesting. I find it funny that it listed a bunch of locomotives with the same tractive effort of 135,375 lbs, including the K4s. I'm sure the PRR wished that was true!

These AI tools have a ways to go, but the progress made with natural language processing and computer language models is impressive.

@GG1 4877 posted:

There was a very interesting story on NPR this morning discussing how AI computing is completely different programming from the computers we are all used to utilizing to do highly complex work.  In short, current AI does cannot accurately predict technical language.

Link to news story is here.

Jonathan, Thanks for the link.  The reviewer's conclusions matched mine that this tool has a long way to go.  Folks should be very wary of its responses and how they use it as it could be dangerous.   

I wouldn't use it on my job, because if there's a problem later and the boss learns I used this artificial tool, he'd kick me out for being too stupid to do so.   

In its present state I'd use it for simply fun tasks and keeping the questions simplistic i.e. recipes,  poems, historical questions (if you know the answer - as I asked a few and it got it wrong), maybe a few modeling related questions - but nothing technical.

@RickO posted:

Interesting, but it kind of defeats the purpose of the 28,720 members on here, no? Surely another human being has information to share.

Rick,

This AI site will take a while for it to get up to speed.  It has a lot of learning to do yet and it is not fully capable of intense technical analysis or recommendation.   

At its present stage its more like a over glorified Google site with the information from several Britannica Encyclopedia's and possibly Wikipedia imbedded into it.

Our combined memberships brain power exceeds this tool - especially when it comes to anything train related.

Thanks for the research - very interesting and informative.  I was unaware that this technology had progressed this far and was publicly available.  As an experiment, I posed my own question...

Me:  What will happen to this thread?

AS (artificial stupidity):  Replies to the thread will shift to focus on political topics and conspiracy theories, and it will be shut down within a day.

Prior to fixing postwar Lionel engines, I wrote digital imaging code for seismic data processing, which is done for oil/gas exploration.  A colleague of mine that is still working emailed me about chatGPT.  He said it wrote the software "for a small cube running around in a large cube (with overlap) such as I need to do with the 5D regularization".  It took him weeks to write the code, and chatGPT did it in minutes.  While it wasn't perfect, it produced an 'amazing skeleton'.

I can see an obvious flaw in the technology compared to how the forum operates today - it doesn't provide opinions or make subjective judgements. Some of the most entertaining posts on train forums come about when two or more experts have differing opinions on what they clearly see as facts. The interchange between humans is safe for the time being. I did a search on "how to grease a ball bearing" and it wasn't nearly as fascinating as a recent thread on the forum.

As an aside, for those who like to read, Kurt Vonnegut wrote a satire in 1952 titled Player Piano that foretold of a world where AI displaced all the workers. The guy was a visionary. As with all of Vonnegut's books it'll get you thinking and laughing at the same time.

I coded in the early days of AI, which were massive if/then.  Very task specific.   AI requires massive data and a good understanding of the process.   Everyone thinks differently and this variances in skill lead to new strengths.   I wonder about Modern AI, how many ghosts are in the machine?

If you really want a SciFi story to make you think, "Food of the Gods", by H.G. Wells.  I think every bio researcher should read it.

@VHubbard,

Our company was a small part of a massive conglomerate, which included large defense contractors.  When the various technical groups from the companies would get together in an attempt for cross fertilization (so to speak), the defense guys didn't say much.  Back in the early 80s they did show an AI system for tank recognition.  As they went through the sanitized demo and described (in non-specific terms) the code I thought, hmmm, I could do that with a lot of if/then/else constructs.  I'm glad that I wasn't too far off the mark, except of course it no longer matters.

@Mallard4468 posted:

Thanks for the research - very interesting and informative.  I was unaware that this technology had progressed this far and was publicly available.  As an experiment, I posed my own question...

Me:  What will happen to this thread?

AS (artificial stupidity):  Replies to the thread will shift to focus on political topics and conspiracy theories, and it will be shut down within a day.

Interesting response from it.  I suspect it has read many of the threads here which quickly were shut down - especially which train manufacturer is the most innovative or has the best track system. 

Attached is a new poem I asked it to generate.  It did so within seconds.

Attachments

Images (1)
  • mceclip0

I've been scouring the web for pictures of the Southern Railway Y6b. I never heard of such an animal but if Silicon Valley's little electronic savant says there was one, it must have existed.

Also, UP can stop crowing about their Big Boy.  According to the machine's answers there were literally dozens of engines that produced 135,375 lbs of T.E.

See GIGO, Garbage In, Garbage Out.

Last edited by Nick Chillianis

AI continues to improve and impress, and chatGPT is no exception. A friend of mine asked it to do a poem about his cat in a particular style, and it succeeded. Though, I’d like to see it be able to apply more complex rules like writing in iambic pentameter or doing a haiku. Given those are just applying some math to words, it should be within its abilities.

I will admit that AI is a little bit scary sometimes. AI art is rather controversial amongst my artists friends, and gaming friends. It’s very good at making “good enough” art for, say characters in role playing games. But it also hurts independent artists commissions. Ultimately though I don’t think it has been as damaging as initially feared, because the quality is not very good and the sorts of people who make use of it weren’t commissioning art anyways. An way to easily spot it is to look at the hands: as any artist will tell you they are a pain to try and draw! Smart AI hides the hands of characters in the posing or subtlety messes them up. Which is amusing because it seems it struggles with the same issues we do.

AI goofs can reveal some amusing things about our collective subconscious. I particularly liked the time an AI after being fed pictures from the Internet decided the difference between a dog and a Wolf was wether it was standing in the snow. Its not wrong, but...

Vonnegut’s Player Piano is definitely a must read. I seem to recall a rather lengthy chapter that takes place on a train, too!

Add Reply

Post
×
×
×
×
Link copied to your clipboard.
×
×