Can AI help us repair our cameras? A test request on my own behalf

Andreas Thaler

Subscriber
Joined
Nov 19, 2017
Messages
4,533
Location
Vienna/Austria
Format
35mm
Artificial intelligence (AI) could also be helpful in finding information about camera repairs.

To get an idea, I asked ChatGPT (App for iOS) for information about me and the Canon T90 (my current focus), as that's the best way to judge the quality.

My search query was

andreas thaler canon t90 service and repair


The result:


Comments
  • It's amazing what AI can quickly gather when querying the web.
  • If I were vain, I'm of course not I would be pleased with this very positive portrayal of myself.
  • But that also makes the whole thing suspicious. Because whether this assessment is accurate or not can only be determined by what I write under my own name and what others might write about me.
  • Whether this makes sense or nonsense is something the AI can't judge, how could it? For that, AI would have to sit down at the desk itself with a screwdriver
  • It reminds me of the days before Google, when metasearch engines were used, querying multiple search engines in parallel to improve search results.
  • The AI doesn't do anything else here except summarize the results in one text, with the sources cited in the app.
  • I find the judgment and tone questionable. What are you supposed to believe if you read this?


Now a search query only on a subject without mentioning a person:

canon t90 eee help


The result:



Comments
  • This makes it clearer that AI cannot yet replace the repair man.
  • What's written here is partly true, but partly not. It's compiled from various web posts, primarily on PHOTRIO. Working through it in this order won't necessarily solve EEE/HELP, but it's close. Provided the information retrieved and compiled from the web isn't nonsense.
  • That is, as long as there's only one source and no contradictions/opposing opinions, the AI itself won't communicate anything different from what it finds.


Conclusion
  • I haven't delved further into AI yet; this was my first foray into a subject area where I know something.
  • The results are helpful, but not miraculous. It's understandable how they're generated and what they depend on.
  • Therefore, I'll use AI to search for information on camera repairs, but I'll approach the results with skepticism and try to verify/falsify them.
  • In any case, AI doesn't do anything on its own—yet.
  • If I can't judge information for truth or falsehood, I won't be successful. And that requires reliable knowledge.
 
Last edited:

mshchem

Subscriber
Joined
Nov 26, 2007
Messages
14,358
Location
Iowa City, Iowa USA
Format
Medium Format
Fascinating, not surprising.

Today I listened while my wife was talking to AI that she thought was human, it was hilarious. Didn't take a minute before the robot gave up and sent her to an actual human.

I understood my dear wife, I've had experience, made sense to me.
 
OP
OP

Andreas Thaler

Subscriber
Joined
Nov 19, 2017
Messages
4,533
Location
Vienna/Austria
Format
35mm

Was it Alexa?
 

Chan Tran

Subscriber
Joined
May 10, 2006
Messages
6,695
Location
Sachse, TX
Format
35mm
Ask ChatGPT this "what is 2343234323*23432943232" and see if it has the correct answer. Microsoft Copilot did it wrong. It gives a close answer but not correct. It's wrong in the 5th significant digit. So not as accurate as an 8 digit calculator.
 

Truzi

Member
Joined
Mar 18, 2012
Messages
2,632
Format
Multi Format
Is the web claiming you are AI? LOL.

Most of the popular AI programs we hear about mention "language models." A large part of this is being able to mimic conversational text (which things were doing well over 15 years ago, but not called AI).

I think it reminds you of the meta-searches because it is very similar - just taken to a higher degree because advances make it possible. It's just a web search that gives you a frequently inaccurate summary narrative instead of only returning links. The "value add" of AI is condensing several sources into blocks of text that seem like someone responded to you in a conversational manner. Someone who has the source text in front of them, but still gets it wrong.

(No it's not!!! It's new, it's shiny, it's different. It's exactly what the marketing tells us it is (despite what experience, observation, and actual tech data from AI firms themselves state.))
 

Kino

Subscriber
Joined
Jan 20, 2006
Messages
7,652
Location
Orange, Virginia
Format
Multi Format
AI is that neighbor who knows just enough to get you in trouble...
 

mshchem

Subscriber
Joined
Nov 26, 2007
Messages
14,358
Location
Iowa City, Iowa USA
Format
Medium Format
Was it Alexa?

Nope, it was an US division of the drug company Eli Lilly. I had overheard at the beginning "It" identify as AI, my wife missed that part and started a conversation, the robot sounded like a human, neutral US accent amazing. My dear wife is an artist and depending on the situation uses my last name, her last name or her legal last name (hers-mine).

As she was telling the Silicon based life for these details Ms. AI started to rock back and forth with smoke coming out of her head and passed the call off to a human.

At this point my wife turned to me and said, "That was AI".

We are going like lambs to the slaughter. As fast as things can change in one's government, the AI thing will be faster.
 

blee1996

Subscriber
Joined
Jul 25, 2008
Messages
1,136
Location
SF Bay Area, California
Format
Multi Format
OP
OP

Andreas Thaler

Subscriber
Joined
Nov 19, 2017
Messages
4,533
Location
Vienna/Austria
Format
35mm

I'm speechless! I'll be using this to write my posts from now on!

In particular, there are no embarrassing praises for people.

Is this available for the iPhone?

I'll have to check right away.

Thank you!
 
OP
OP

Andreas Thaler

Subscriber
Joined
Nov 19, 2017
Messages
4,533
Location
Vienna/Austria
Format
35mm
You just have to ask Perplexity the right questions!

Summary:
Andreas Thaler is a respected voice in the Canon T90 community, offering practical advice on buying, maintaining, and repairing this iconic camera, which remains a favorite among film enthusiasts for its build quality and advanced features.

But that only happens when I enter my name + Canon T90.

But how does the AI come to such a conclusion? It can't measure the quality of what I write.

That interests me; I'll have to enter the namens of my friends right away to see if they're also respected voices
 
OP
OP

Andreas Thaler

Subscriber
Joined
Nov 19, 2017
Messages
4,533
Location
Vienna/Austria
Format
35mm
Ask ChatGPT this "what is 2343234323*23432943232" and see if it has the correct answer. Microsoft Copilot did it wrong. It gives a close answer but not correct. It's wrong in the 5th significant digit. So not as accurate as an 8 digit calculator.

There's probably still a Pentium processor working there that has been overseen
 

BHuij

Member
Joined
Oct 12, 2016
Messages
812
Location
Utah
Format
Multi Format
I work in data engineering and analytics. The number of people who think AI is the future of software engineering (or writing code of any sort, even SQL queries or data models) is frightening. Given a sufficiently good foundation, Claude and even ChatGPT can save some time writing simple but tedious queries or scripts, provided you're willing to properly evaluate and QA the output to make sure it's actually doing what you want. In my case, I'd rather write the code myself than essentially debug code written by "someone else" about 9 times out of 10.

But as professionals who know what they're doing move out of the workforce, and the source material used to train these large language models becomes increasingly cannibalistic (AIs learning to code by ingesting bad code spit out by other AIs), we're going to end up in a situation where software companies have entire code bases that were written by AI, have fundamental problems, and nobody is left who knows enough to fix them.
 

Chan Tran

Subscriber
Joined
May 10, 2006
Messages
6,695
Location
Sachse, TX
Format
35mm
There's probably still a Pentium processor working there that has been overseen

Out of curiousity I argued with it and said that it was wrong. Found out it did the calculation using scientific notation and it did right but when converting to standard notation it made mistake. I don't know how it changed the mantisa when it did the conversion.
 

Truzi

Member
Joined
Mar 18, 2012
Messages
2,632
Format
Multi Format
I had a discussion with ChatGPT about The Great Tomsoni - someone I knew a little about because I was into magic back in high school.

It got nothing right, except that he was a Magician. A google search will return the correct result in the first hit.
ChatGPT kept apologizing, telling me I was correct, adding more incorrect information in the process. The browser session eventually crashed, lol.

The only reason I had even done that is because I was in Security training as part of my IT job, and the consultant who came in kept referring to ChatGPT to get answers to example questions. GPT was often incorrect (using common sense) and if we searched actual web pages we'd find the correct answer. Still, the consultant, who had been doing IT since the 90s, often deferred to the chatbot.
 

BHuij

Member
Joined
Oct 12, 2016
Messages
812
Location
Utah
Format
Multi Format

All these AIs are large language models, not large math models. The process used to train them is ultimately about how effectively they return an output that the user likes or agrees with, with little regard for whether or not it's factually accurate. Now of course, most users will have a more favorable response to a factually accurate response than a factually inaccurate one. But asking a large language model to do math problems is like asking an automatic calculator to write a novel. It's just not the best tool for the job. Doesn't mean it won't get things right frequently, just that it has a high probability of confidently giving you an answer that is "kinda close and definitely believable but ultimately wrong."
 
OP
OP

Andreas Thaler

Subscriber
Joined
Nov 19, 2017
Messages
4,533
Location
Vienna/Austria
Format
35mm
Where I rely on AI is translating my German posts here into English.

Google Translate does that, and I intervene where I think it's necessary.

It's quick, no one has complained yet, and I suspect it's US English that Google returns to me.
 

BHuij

Member
Joined
Oct 12, 2016
Messages
812
Location
Utah
Format
Multi Format
Where I rely on AI is translating my German posts here into English.

Google Translate does that, and I intervene where I think it's necessary.

It's quick, no one has complained yet, and I suspect it's US English that Google returns to me.

Even before AI, translation algorithms between major languages were really quite good. I think making them even better is a perfect use case for LLMs.
 

Truzi

Member
Joined
Mar 18, 2012
Messages
2,632
Format
Multi Format
Where I rely on AI is translating my German posts here into English.

Google Translate does that, and I intervene where I think it's necessary.

It's quick, no one has complained yet, and I suspect it's US English that Google returns to me.

If the translation makes sense, it's probably not US English, lol.
 

Sharktooth

Member
Joined
Apr 4, 2022
Messages
322
Location
Canada
Format
Medium Format
Wow! Perplexity is very useful. I just did a search for "Kodak Retina IIIc exposure counter repair". It provided quite a bit of useful information, and also found a link here in Photrio where I asked the same question a while ago. The folks here at Photrio provided the answer for me.

This was infinitely better than a Google search.
 

blee1996

Subscriber
Joined
Jul 25, 2008
Messages
1,136
Location
SF Bay Area, California
Format
Multi Format
Yes I agree: Perplexity has done well with combining search and AI summarization, while reducing hallucination (making up stuff). For some use cases, Perplexity can do better than Google Search.

We still need to cross-verify what AI says, since hallucination is inevitable (even though can be minimized).
 
Cookies are required to use this site. You must accept them to continue using the site. Learn more…