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

[mshchem]

View attachment 399145

View attachment 399144

I have long been receiving my orders from HAL, who speaks to me through my cameras

YIKES!!!!

 

[Andreas Thaler]

But if the AI lives off other people's texts, it can't improve them either. Junk remains junk. At least the sources are displayed.

 

[Andreas Thaler]

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

For me, as a native German speaker, textbooks by US authors are usually easier to read than those by British authors. It seems to me that more emphasis is placed on simplicity. But that could be due to my vocabulary.

 

[koraks]

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.

Indeed. But it won't matter as much as you think, because our approach towards building and maintaining software will also change. Presently, if something doesn't work, we debug it. In the future, we won't, because (1) it won't be possible for the reason you point out, and (2) it won't be economical due to labor costs. Instead, we'll ditch the broken/buggy module and ask AI to produce a new one. The logical next incremental step is of course that we'll just have AI spit out 128 variants of the same model and then have another AI tool evaluate them against our functional requirements, so we select the one we like the most. My expectation is that we'll shift the focus of human intelligence in software engineering from code production to architectural design. An area where SE seems to be lagging behind to begin with, so maybe it's about time, too.

 

[loccdor]

I asked AI a year ago about the history of Fuji films and there was no information on the English web about them so all its information was wrong. So I ran Fuji's Japanese site through a translator and used it to produce an English language Wikipedia article. Now when I ask AI about Fuji films history, it gets all its information from that Wikipedia article that I wrote.

It's a real problem, I could easily have injected any misinformation I wanted and it would be reporting it as fact. It's only because I didn't do that it had any accuracy at all.

 

[koraks]

It's a real problem, I could easily have injected any misinformation

Even worse, 'your' wiki entry may be riddled with errors that you couldn't spot because you relied solely on an AI interpretation of the source material.

 

[Andreas Thaler]

Referring to my initial question, I can summarize that Perplexity, at least for me, is useful when researching repair projects.

The app queries the sources it cites in parallel and summarizes the results meaningfully.

The quality of the result depends on the quality of the sources, meaning I have to validate the results—as with all research.

In practical terms, this means I'll be using Perplexity in conjunction with Google.

 

[loccdor]

Even worse, 'your' wiki entry may be riddled with errors that you couldn't spot because you relied solely on an AI interpretation of the source material.

I didn't use AI, I used a translator, and it was mainly information about dates certain films came out so there wasn't all that much room for error.

 

[koraks]

I didn't use AI, I used a translator, and it was mainly information about dates certain films came out so there wasn't all that much room for error.

OK, sorry, my wording was sloppy. I referred to the translator as 'AI' as well. My apologies. And this wasn't to disqualify your wiki article, which undoubtedly is useful and I trust you were diligent in writing it. What I intended to demonstrate is the inherent problem of 'unk-unks' (unknown-unknowns) or systematic uncertainty in using AI. It creates a situation where we may be blissfully unaware of how we're being led astray.

 

[loccdor]

@koraks No need to apologize, I appreciate your views and contributions! And I agree that this technology is going to spiral us into a feedback loop of sorts. Being simultaneously helpful and corrosive. Just a new landscape everyone is going to have to navigate, for better or worse. I wish us all luck in that.

 

[koraks]

Yeah - it's a brave new world alright!

 

[Chan Tran]

Funny and it's true. I asked Copilot where Fujifilm made their new X Half camera and it doesn't know. Where things are made is the question very hard to find the answer.

 

[Alan Johnson]

I tried Perplexity.ai for Canon AE1 stuck shutter and received some apparently good instructions.

 

[F4U]

Every time I do a search the first one in the listings is an AI answer. I ignore it. Totally worthless. Artificial intelligence. That describe some people I know, and ones like that often are involved in the programming.

 

[Andreas Thaler]

Sources!

AI can't do it on its own if there's no solid information to learn from.

In my research on repair topics online, I notice that people usually ask for videos for solutions. As soon as text is mentioned, there is silence. There are rarely requests for books or technical documents at all. A quick fix seems to be important.

But that rarely works; the (newer) cameras and their problems are too complex for that.

The Camera Craftsman, SPT Journal, and C & C Guides provide in-depth information that, of course, must be accessed. Most of the documents have been available for a long time via Learn Camera Repair, for which I am very grateful to founder Gene Pate.

Accordingly, these sources do not appear in my AI research. Instead, in some DIY forums, you can find nasty references to old knowledgeable nerds lurking in old fashioned text based forums who are at least useful when it comes to solving problems that a video doesn't provide.

I don't understand the claim. Why don't DIY enthusiasts get involved? Are they afraid of the effort?

It seems that only easily consumable information finds a buyer. But that won't help our aging pets.

And AI won't be able to either. Because, like humans, it can't predict the future. It needs reliable information to learn. And if that's not available, it can't produce anything useful.

At least not tomorrow.