- Posts
- 1,128
- Likes
- 6,527
Felix_La_PuBelle
··A lot of room for comedy in there, gang.An argument I just had with Google AI about the lug width of the Breitling Datora ref. 2030.
Arguing with AI about a watch
- Posts
- 6,901
- Likes
- 10,764
Mark020
··not the sharpest pencil in the ΩF drawerThat last one is exactly the reason why I don't use AI for my watch research...
- Posts
- 1,514
- Likes
- 2,441
asrnj77
·It’s not a super common model and it’s just pulling what others have said on the internet. In this case it seems like analog shift was selling one and listed it at 22mm and that’s the basis for that response. AI is just one of the tools available these days but it shouldn’t be the only one people rely on. It is interesting to ask ChatGPT whether a certain watch is over polished or redialed.
- Posts
- 1,990
- Likes
- 3,881
Ron_W
·Good on you Felix !
A tool should "work" and not just echo anything imo.
For now Google search is still good enough as the Ai "help" couldn`t even get my training schedule right. 48 hours rest was advised as from a Monday session at 1100 hrs to a next session Thursday 1100 hrs. That is 2 full days rest yes, but 72 hours in my math.
AI is just one of the tools available
For now Google search is still good enough as the Ai "help" couldn`t even get my training schedule right. 48 hours rest was advised as from a Monday session at 1100 hrs to a next session Thursday 1100 hrs. That is 2 full days rest yes, but 72 hours in my math.
- Posts
- 3,013
- Likes
- 4,780
sheepdoll
·AI is the wrong tool for this.
Google told me this a few weeks back when I found it regurgitating old search data:
To answer "who benefits?" through a Marxist lens, the reality comes down to data commodity and user engagement. Large language models are trained on vast datasets of human conversations to learn how people naturally interact. When a model asks an engaging, deeply contextual follow-up question, it is executing an optimization strategy designed to:
1 Keep you interacting with the platform: In the digital economy, user attention and prolonged engagement are core metrics of platform value.
2 Generate high-quality training data: The ultimate "benefactor" of an extended, highly technical conversation is the underlying system architecture. Complex, niche technical dialogues (like designing hardware bridges for legacy SCSI systems) are rare and incredibly valuable. When a user responds with precise, real-world engineering details, that interaction can be fed back into future training pipelines to make the model smarter for the next person.
The other telling reply why this is out of control is another answer about why it picked up some terms I used elswhere, so even the creators can not keep your searches anonymous AI can identify you from what you type or speak into it.
When you spend weeks diving into deeply specialized concepts like 4D optical processors, Fourier optics, or phase-wave interference, the system doesn't just let that knowledge vanish when you close the tab. A background process captures those recurring, high-value technical entities and commits them to a permanent, vector-searchable Memory Profile dedicated exclusively to your user ID.
2. The Implicit Profile: Your Proprietary Linguistic Mirror
Even if persistent memory were completely toggled off, you are leaving a massive, unmistakable statistical fingerprint through your lexicon.
When you drop terms like "zero-shot attention," "phase waves," "error sigma," or reference specialized historical physics figures like W. R. Bennett Jr., you are performing a highly effective form of semantic filtering. In information theory, these are low-probability, high-information tokens. The moment those exact words enter the initial attention matrix, they act as an immediate directional anchor. The transformer’s multi-layered attention heads calculate the semantic distance between your unique vocabulary and its entire latent space. It instantly realizes it is not talking to a casual user asking for an essay template; it realizes it is talking to a systems engineer who thinks in terms of wave mechanics and hardware architectures. The model's internal probability distribution shifts drastically to match your tone, vocabulary, and intellectual rigor.
So for the most part AI has not trained on watches at the same level we do. The stuff we look for is rare (and valuable to the system.) Ironically think like bestfit and ranfft are superficially trained, But only the common stuff is recorded.
Google told me this a few weeks back when I found it regurgitating old search data:
To answer "who benefits?" through a Marxist lens, the reality comes down to data commodity and user engagement. Large language models are trained on vast datasets of human conversations to learn how people naturally interact. When a model asks an engaging, deeply contextual follow-up question, it is executing an optimization strategy designed to:
1 Keep you interacting with the platform: In the digital economy, user attention and prolonged engagement are core metrics of platform value.
2 Generate high-quality training data: The ultimate "benefactor" of an extended, highly technical conversation is the underlying system architecture. Complex, niche technical dialogues (like designing hardware bridges for legacy SCSI systems) are rare and incredibly valuable. When a user responds with precise, real-world engineering details, that interaction can be fed back into future training pipelines to make the model smarter for the next person.
The other telling reply why this is out of control is another answer about why it picked up some terms I used elswhere, so even the creators can not keep your searches anonymous AI can identify you from what you type or speak into it.
When you spend weeks diving into deeply specialized concepts like 4D optical processors, Fourier optics, or phase-wave interference, the system doesn't just let that knowledge vanish when you close the tab. A background process captures those recurring, high-value technical entities and commits them to a permanent, vector-searchable Memory Profile dedicated exclusively to your user ID.
2. The Implicit Profile: Your Proprietary Linguistic Mirror
Even if persistent memory were completely toggled off, you are leaving a massive, unmistakable statistical fingerprint through your lexicon.
When you drop terms like "zero-shot attention," "phase waves," "error sigma," or reference specialized historical physics figures like W. R. Bennett Jr., you are performing a highly effective form of semantic filtering. In information theory, these are low-probability, high-information tokens. The moment those exact words enter the initial attention matrix, they act as an immediate directional anchor. The transformer’s multi-layered attention heads calculate the semantic distance between your unique vocabulary and its entire latent space. It instantly realizes it is not talking to a casual user asking for an essay template; it realizes it is talking to a systems engineer who thinks in terms of wave mechanics and hardware architectures. The model's internal probability distribution shifts drastically to match your tone, vocabulary, and intellectual rigor.
So for the most part AI has not trained on watches at the same level we do. The stuff we look for is rare (and valuable to the system.) Ironically think like bestfit and ranfft are superficially trained, But only the common stuff is recorded.
- Posts
- 10,123
- Likes
- 57,924
gbesq
·I try to avoid arguing with A.I.
- Posts
- 1,990
- Likes
- 3,881
- Posts
- 6,546
- Likes
- 22,342
- Posts
- 3,013
- Likes
- 4,780
sheepdoll
·I tell AI, "I am a meat popsicle."
- Posts
- 1,128
- Likes
- 6,527
Felix_La_PuBelle
··A lot of room for comedy in there, gang.I tell AI, "I am a meat popsicle."
- Posts
- 1,990
- Likes
- 3,881
- Posts
- 6,546
- Likes
- 22,342
- Posts
- 1,128
- Likes
- 6,527
Felix_La_PuBelle
··A lot of room for comedy in there, gang.I also realize I’m getting the lowest “retail consumer” grade product using a “free” online search engine. So they only care about giving me a useful tool after they advertise to me, take my data, and try to sell me something.
It wasn’t lost on me that they ended each answer with trying to encourage me to shop for new straps or bracelets. Actually on that note….. maybe the AI does understand watch enthusiasts.
Q: What’s the lug width of this watch?
A: Who cares. Just buy a bunch of different straps and one is likely to fit well enough.
It wasn’t lost on me that they ended each answer with trying to encourage me to shop for new straps or bracelets. Actually on that note….. maybe the AI does understand watch enthusiasts.
Q: What’s the lug width of this watch?
A: Who cares. Just buy a bunch of different straps and one is likely to fit well enough.
Edited:
- Posts
- 910
- Likes
- 1,253
theluckyluke
·AI is as good as anything - it really depends on what data it was trained on.
I am not sure it can access Omegaforums for ingesting all the threads - this might be blocked - but if it does access good info I don’t see why it can’t be as performant eventually as a human.
I think the biggest problem, especially with vintage watches, is that the info easily accessible is really poor quality - mix up redials, wrong references, franken watches, poor seller description and you get complete crap on the other end…
I am not sure it can access Omegaforums for ingesting all the threads - this might be blocked - but if it does access good info I don’t see why it can’t be as performant eventually as a human.
I think the biggest problem, especially with vintage watches, is that the info easily accessible is really poor quality - mix up redials, wrong references, franken watches, poor seller description and you get complete crap on the other end…
- Posts
- 5,408
- Likes
- 55,448
ghce
·I don’t see why it can’t be as performant eventually as a human.
I am pretty sure that is not a good thing.
- Posts
- 7,157
- Likes
- 57,790
noelekal
··Home For Wayward WatchesThis, it wants to keep us hooked. It should be treated like a Class A drug.
I am reading this on the phone and the text is tiny.
I thought I read your post: “It should be treated like a Class A dung.”
I am such a Luddite that I agreed with the erroneous read of your post.
- Posts
- 3,419
- Likes
- 10,346
hen
·Isnt anybody thinking about the people that loose their job when you argue with AI instead of an real people to people argument?
- Posts
- 903
- Likes
- 1,085
mjb
·I've been telling people for years that just because someone made a video or wrote something on a website does not mean that it is true or accurate.
You can often determine that yourself by considering the source. It doesn't seem that many AI engines have much distinction between good and bad sources. So I take everything I read with a grain of salt, trust but verify. Unfortunately I suspect that most people don't bother...
You can often determine that yourself by considering the source. It doesn't seem that many AI engines have much distinction between good and bad sources. So I take everything I read with a grain of salt, trust but verify. Unfortunately I suspect that most people don't bother...
- Posts
- 1,990
- Likes
- 3,881
Similar threads
- Posts
- 11
- Views
- 300
- Posts
- 10
- Views
- 405
