Hermès in the Digital Age: ChatGPT AI responds to Hermès trivia

TPF may earn a commission from merchant affiliate
links, including eBay, Amazon, and others

R

r luvs h

Member
May 29, 2022
1,186
5,152
#1
Silly post incoming:
As AI (artificial intelligence) technology becomes more and more developed, I thought I would dip my toes in the AI pool to see what all the fuss was about. I decided to try ChatGPT— an AI program created by Open AI that can respond to an impressively wide variety of questions, no matter how seemingly niche or esoteric.

For my experiment, I decided I would grill ChatGPT with some Hermès trivia ranging from well-known facts to some more obscure information in order to see how the AI responded. I was quite impressed with my findings! Granted, some of the responses were completely and hilariously incorrect, but others were surprisingly accurate. I was most impressed with the AI’s ability to unlearn and relearn information, upon my correction. The following is some of my dialogue with Chat GPT.

To the prompts, “Tell me about Hermès Epsom leather” and “Tell me about Hermès Clemence leather,” this was the AI’s response:

epsom.pngclemence.png

Next, I asked about the discontinued leather Courchevel:

courchevel.png

I then decided to increase the difficulty by asking a question that required a more detailed response: "What leathers replaced Hermès Courchevel and Gulliver?" Unfortunately, the AI responded incorrectly to my prompts, so I had to teach it the correct answer.

courchevel + gulliver 1.pngcourchevel + gulliver 2.png

Impressively, the next time I asked this question, the AI responded correctly:

courchevel + gulliver 3.png
 
Last edited:
  • Like
  • Wow
  • Haha
Reactions: Holsby, alovefor, JavaJo and 10 others
#2
Then I decided to ask some harder questions about colours and their debut years, and the AI’s responses were not correct. It made up a non-existent colour called “Flame”, confused the debut year of Crevette with Sanguine, and did not know what pink shades were introduced in 2006. These are harder questions, so I was not surprised.

%22flame%22.png
(The correct answer to my question is "Terre Battue".)

X sanguine.png
(The correct answer to my question is "Crevette".)

2006 pinks 1.png2006 pinks 2.png

I corrected some of the AI’s responses, and it quickly learned the correct answers. It even added some of its own, factual flare!

2006 pinks 3.png

All in all, I was rather impressed by the AI’s knowledge and ability to create coherent responses (even when those responses were incorrect). I feel like ChatGPT could be very useful for those trying to understand rudimentary information about Hermès. I also feel like, with some teaching and reprogramming, ChatGPT could be very helpful to those more versed in Hermès for quickly recalling information like production years for certain colours, leathers, and styles.

And of course, after I finished quizzing the AI, I asked for it to spare me when it inevitably becomes sentient and tries to take over the world— I am not sure I like its response.:lol:

final words.png
 
  • Haha
  • Like
Reactions: Holsby, alovefor, JavaJo and 10 others
#8
Just for fun, some more AI facts….As someone who works with AI every day, know that the writing of prompts is what defines the response. One thing that both language and image generative AI has a hard time with is anything in a foreign language (thus CGPT3 thinking Feu is Flame and translating Sanguine halfway). Anything non-English tends to throw it off as it attempts to translate.

You can test before chatting if an AI has been “trained” on the material you want to hear about, with some basic questions (like the ones asked). Clearly this particular chatbot has had only partial training around Hermes (could be annual reports since those are used a lot in training).
Remember - it doesn’t scrape the web - it has been taught from specific materials and is constantly learning. Now it’s been taught some H tricks!
 
  • Like
  • Wow
Reactions: Holsby, lovejoysunshine, miss mitzi and 3 others
#9
QuelleFromage said:
Just for fun, some more AI facts….As someone who works with AI every day, know that the writing of prompts is what defines the response. One thing that both language and image generative AI has a hard time with is anything in a foreign language (thus CGPT3 thinking Feu is Flame and translating Sanguine halfway). Anything non-English tends to throw it off as it attempts to translate.

You can test before chatting if an AI has been “trained” on the material you want to hear about, with some basic questions (like the ones asked). Clearly this particular chatbot has had only partial training around Hermes (could be annual reports since those are used a lot in training).
Remember - it doesn’t scrape the web - it has been taught from specific materials and is constantly learning. Now it’s been taught some H tricks!
Click to expand...
Very true! I should add that this version of ChatGPT was last updated in 2021, so anything that has happened since then will not be in its knowledge database.

I wonder if it only has problems with English-French mixture? I tried communicating with the AI in a few other languages like Greek and Latin, and it seemed to perfectly comprehend and respond to everything I said. I also played a couple rounds of 20 questions in Spanish with the AI, which was rather entertaining. Even when I used colloquialisms and slang terms, the AI had no issues responding to my prompts and guessing correctly each time.
 
  • Like
Reactions: Holsby
#11
r luvs h said:
Very true! I should add that this version of ChatGPT was last updated in 2021, so anything that has happened since then will not be in its knowledge database.

I wonder if it only has problems with English-French mixture? I tried communicating with the AI in a few other languages like Greek and Latin, and it seemed to perfectly comprehend and respond to everything I said. I also played a couple rounds of 20 questions in Spanish with the AI, which was rather entertaining. Even when I used colloquialisms and slang terms, the AI had no issues responding to my prompts and guessing correctly each time.
Click to expand...
It’s not that the AI doesn’t understand other languages; it’s that the use of them can confuse prompts.
 
#12
For fun, I decided to re-ask a couple of @r luvs h 's questions. You can see that CGPT3.5 doesn't learn everything it's told by users (which is good, because it's not supposed to - only OpenAI is supposed to be able to train it, otherwise people could fill it with misinformation).
I am using the current version of ChatGPT, though - this is ChatGPT3.5 which is updated constantly (there is also ChatGPT4). I am not certain that there is a version that has not been updated since 2021?

Screen Shot 2023-04-29 at 8.12.54 AM.png

And, because this question is asked so often, I figured I'd see what CGPT3.5 had to say :lol:. Again, not entirely accurate:


Screen Shot 2023-04-29 at 8.20.52 AM.png
 
  • Like
  • Haha
Reactions: Holsby, PACNWPurseLover, shrpthorn and 2 others
#13
Does Hermes conduct their own auctions? Nope. I saw a 60 minutes episode about Ai and forsee a huge need for oversight to correct incorrect information. AI re: Hermes gathers its information from the internet, which is full of misinformation and regurgitates the information it is exposed to. It is a tremendous asset if the information is reliable.
 
Last edited:
#14
haute okole said:
Does Hermes conduct their own auctions? Nope. I saw a 60 minutes episode about Ai and forsee a huge need for oversight to correct incorrect information. AI re: Hermes gathers its information from the internet, which is full of misinformation and regurgitates the information it is exposed to. It is a tremendous asset if the information is reliable.
Click to expand...
No, it does not scrape the internet (it actually will not do this even if you ask it to). It is trained on specific materials, which may very well be incorrect and will almost always be incomplete.

This is just a fun game here; no one should be getting information from a non-professional AI and acting on it. There are trained AIs with serious specialties (such as reading radiology scans) which are more useful - but all of them are exactly as good as what was input into them. We call this GIGO - garbage in garbage out.
 
  • Like
  • Love
Reactions: Holsby, alovefor, shrpthorn and 1 other person
#15
QuelleFromage said:
No, it does not scrape the internet (it actually will not do this even if you ask it to). It is trained on specific materials, which may very well be incorrect and will almost always be incomplete.

This is just a fun game here; no one should be getting information from a non-professional AI and acting on it. There are trained AIs with serious specialties (such as reading radiology scans) which are more useful - but all of them are exactly as good as what was input into them. We call this GIGO - garbage in garbage out.
Click to expand...
I admittedly know absolutely nothing about AI and its knowledge of Hermes. My impressions of AI came from Sundar Pichai, CEO of Google and the information he detailed on 60 minutes. While we know better, there are many less experienced who could rely heavily on the answers given. BTW, Congrats on your unicorn Bolide, would love a modeling pic!
 
You must log in or register to reply here.
Top