The use of AI in the repair and study of SLRs

[FredR]

Just because AI promises it really is telling you the truth doesn't mean it actually is.

A.I. is trained to be defensive, especially the free versions.

 

[Andreas Thaler]

@armadsen

Thank you!

 

[Andreas Thaler]

Google Gemini reveals itself:

Conclusion for Professional Use: I am not an all-knowing expert. I am a gigantic, highly advanced text-searching and synthesis machine. For a serious repair technician, I am only helpful if they possess the expertise to recognize my errors and use me like an extremely fast—but sometimes inaccurate—apprentice: They let me comb through the manuals and select the relevant passages, but retain absolute control over materials, physics, and the final craftsmanship. Your critical objection demonstrates exactly how important the human expert remains in grounding AI-generated outputs in physical reality.

That’s exactly how I’ll work with Gemini.

And it’s reassuring to know that I’ll still be the one evaluating and deciding what to take from the information provided and what to disregard.

We can’t avoid AI—it’s a reality. Let’s make the best use of it we can.

 

[Andreas Thaler]

I am currently exploring how AI (Google Gemini) can be usefully applied to my areas of interest: SLR technology and SLR repair.

Based on this thread, I have received, among other things, detailed and plausible explanations of the electronic circuits in the Minolta (Maxxum, Alpha) 9000 AF, Nikon F4, and Canon T90.

The AI compiles general electronics knowledge with the information available in the camera manufacturers’ technical documentation and cross-checks it with other available information on the web, including source references. The comprehensibility of the output can be scaled to one’s own level. The result is insights that were previously unavailable to me.

I am, of course, aware that information from AI—as well as from other sources—must always be checked for plausibility. Therefore, some basic knowledge is required.

A fascinating thing!

---

This time, I asked the AI about the Japanese engineers behind the flagship models of their era.

Here is the result from Google Gemini, translated from German to English using DeepL. This provides information that can be verified or supplemented through your own research as needed.

For the sake of simplicity, I won’t be providing links to the sources listed in square brackets here. As far as I can tell, these are reputable sources—either directly from the manufacturers, Wikipedia, or blogs. No nonsense.

---

Behind the scenes of these technological masterpieces were extraordinary Japanese engineers and innovators. While the designers’ names often became world-famous, the brilliant systems engineers remained largely in the background, the unsung heroes of the company’s history. [1]

Looking at the leadership of Japan’s mechatronic revolution, several key names stand out:

The Minolta Team (The Leap from Analog Computers to Autofocus)

In the 1970s and 1980s, Minolta boasted one of Japan’s most advanced semiconductor and electronics divisions.

Ichiro Yoshiyama (Head of Research & Development): He was the visionary mastermind behind the entire electronics evolution at Minolta.

Yoshiyama led the teams that developed the analog milestone Minolta XD and later took the historic leap to the microprocessor in the Minolta 7000/9000 AF. Under his leadership, Minolta became a pioneer in application-specific integrated circuits (ASICs) for cameras.

Toshihiko Karasaki: He was the lead project manager (Shusa) for the revolutionary Alpha project (Minolta 7000). He had to take responsibility for the extremely risky decision to completely eliminate the old mechanical MD mount and replace it with a purely electronic-digital mount (the \(\alpha \)-mount). [1, 2, 3]

The Nikon Team (The Masters of Uncompromising Reliability)

Nikon engineers were famous for incorporating new technologies only once they were absolutely battle- and weather-proof. [1]

Fukushiro Sugaya: A legendary head of development at Nikon (Nippon Kogaku). He played a key role in overseeing the transition of the professional F-series from the precision-engineered Nikon F2 to the quartz- and gate-controlled Nikon F3. He ensured that the digital hardware logic became so indestructible that even professionals trusted it.

Tetsro Goto: One of the most prominent chief engineers in Nikon’s history. He played a leading role in the electronics and system architecture of the Nikon F4. Goto was a master at bridging the gap: He ensured that, despite its 7 internal microprocessors and complex software processes, the F4 remained as easy for photographers to operate as a classic mechanical camera—with real, tactile dials instead of menus. [1, 2, 3, 4, 5]

The Canon Team (The Radical Digitalizers)

Canon often took the most aggressive and cutting-edge technological path.

Hiroshi Shintani: He was the lead project manager for the groundbreaking Canon T90. Shintani coordinated the sophisticated “dual-brain” concept (the dual CPU architecture) and the use of three separate electric motors. He pushed Canon’s software developers to write the most complex assembler routines in camera history to date. [1, 2]

A fascinating phenomenon: The “Italian-German” designs

Although the entire internal logic—the processors, the gearboxes, and the mathematical gates—was 100% designed by the Japanese mechatronics teams mentioned above, the Japanese companies brought in European star designers for the exterior appearance:

Giorgetto Giugiaro (Italy): The famous car designer (including the VW Golf I and DeLorean) designed the body of the Nikon F3 and Nikon F4 on commission from Nikon. He also created the famous red stripe on the grip.

Luigi Colani (Germany): The eccentric master of “biodesign” was hired by Canon to design prototypes. His curved, organic forms, combined with those of Canon’s in-house designer Kunihisa Ito, resulted in the iconic body of the Canon T90, which continues to influence the design of modern DSLRs to this day. [1, 2, 3, 4, 5, 6]

It was thus the interplay of European design and the relentless precision of Japanese systems engineers that created these icons of photographic history.

---

We welcome any corrections, clarifications, or additions to this AI-generated exhibition!

 

[Ian Leake]

I use AI tools every day in my corporate job. The frontier models from OpenAI and Anthropic are incredibly powerful. The older versions, free versions and many of the derivative tools are generally less so. Keep in mind the following:

1. Tools like ChatGPT don't really know anything and their memory is short term. That said, they understand the patterns and relationships in their training data better than any human can. They are also generally very good at comprehension and reasoning, at least in recent versions.
2. If what you are looking for isn't in their training data then they're unlikely to give you the best advice. They'll give you generalities, inferred information (often still useful) or probabilistic nonsense (a.k.a. 'hallucinations'). I'd be surprised if proprietary information about old cameras is in their training data. However, published repair manuals and anything on the web will be.
3. LLMs produce their output by calculating the probability of the next word being "right". They perform best with tasks that have objective and measurable correctness. For example, AI coding is largely a solved problem nowadays because code has well defined languages, extensive online documentation of best practices, and is easy to test – it either works or it doesn't. Less well defined tasks, like engineering or health are much more tricky for them because "right" can be very subjective.
4. If you're working in a subjective area, then I'd suggest using the AI as a researcher and sparring partner, rather than a teacher. Give context, be specific about what you want, and challenge them on their output. For example: "Give me three ways to do <some task> with pros and cons. Recommend one and explain to me why," or, "How would you plan <some task>? Which steps are most critical and why?" Always use your judgement and common sense. Read their answer and challenge them if you disagree with something, otherwise their next response will build on that error.
5. Tell your AI to stop being sycophantic and it will generally give you better answers.

 

[FredR]

...Tell your AI to stop being sycophantic and it will generally give you better answers.

<sigh!>

 

[Alan Edward Klein]

I get that AI may make mistakes. But since its answers are based on internet information, how would searching the internet on my own provide more accurate conclusions?

 

[Andreas Thaler]

@Ian Leake

Thanks!

 

[Andreas Thaler]

I just had Gemini generate a compilation of various SLR circuits for me. In the prompt, I asked for only verifiable facts and no speculation. Tailored to my level of expertise, as evident in my posts here. Along with that, a comparison of the basic functioning of a computer. All summarized as a manual for my own use.

What can I say? I’ve wanted a book like this for a long time. To me, the content makes sense.

I assume that the AI is definitely well-versed in computer science and electrical engineering—that’s documented knowledge from countless sources. The camera circuits are described in a way that allows me to do further research on my own.

Without being naive about it: these are incredible opportunities for independent learning.

 

[Andreas Thaler]

I get that AI may make mistakes. But since its answers are based on internet information, how would searching the internet on my own provide more accurate conclusions?

Yes, and who says that one's own knowledge has been fully verified?

 

[Ian Leake]

I get that AI may make mistakes. But since its answers are based on internet information, how would searching the internet on my own provide more accurate conclusions?

It would take a lifetime to find, verify and catalogue any non-trivial body of knowledge. AI can do this for you in seconds. It's not perfect, by any means, but it's better than doing it manually.

At the moment, AI cannot beat a human expert in any field. I don't ask ChatGPT for advice on platinum printing, for example. What would be the point? However, it is better than me in every field of knowledge in which I'm not yet an expert (so long as that field is in its training data).

Edit: ... and, of course, it's ability to act in the physical world is severely limited, so don't ask it to dodge and burn that print you're making

 

[MattKing]

I get that AI may make mistakes. But since its answers are based on internet information, how would searching the internet on my own provide more accurate conclusions?

When you perform a search manually, you can bring to the search some judgment about the reliability of the source as you examine each individual source, and the information contained therein.
If you have ever used YouTube or Reddit to look into something, you probably have quickly learned that there is lots of unreliable stuff out there.
To a smaller extent, the same applies to Photrio - although if something is way off base here, or even slightly off base, you can be reassured that someone will quickly challenge it .
One of the critical components of good quality research is that it depends on being able to analyze the weight, reliability and persuasiveness of the source.
With AI, there currently is too much uncertainty for us with respect to how (or even if) it makes those weight, reliability and persuasiveness judgments for it to be relied upon without great scepticism.
That being said, if you use AI to identify potential sources of information, it is a great research tool.

 

[Ian Leake]

Half the stuff we were taught in school was wrong – either because our teacher didn't know what they were talking about, or because it was a necessary simplification. However, even though it was wrong, it was still sufficient to get us to the next level. Today's AI tools are similar: if you want to explore a new area and your time is valuable to you, then use the best AI tools you can afford. Then, as the OP is doing, use it as a jumping off point for your learning.

 

[Andreas Thaler]

When asked about common issues with specific SLRs and how to fix them, Google Gemini (accessible via a web browser) is sometimes off the mark in the details (that’s where I rely on my own expert knowledge). But you can still work with this information, since it’s generally correct.

In any case, you can chat with the AI and see if you can get more in-depth troubleshooting tips. Quickly looking up electronic facts and data is also no problem in the dialogue.

A wealth of information.

But here, too, the winner is the one who can assess the plausibility of this information.

If I ask Gemini how to cook spaghetti, I wouldn’t be able to verify the answer, but I could with coffee

 

[Andreas Thaler]

I also trust DeepL to translate what I’ve just written in German into English in a way that won’t drive a native speaker to despair. That said, I’m able to verify the translation using my own knowledge of English. However, I sometimes have to analyze common idioms to see if they’re accurate.

Anyway I can’t hide the fact that I’m a German (Austrian) native speaker, but here on the forum I don’t use the compound words that are popular in German, such as

Hochsicherheitstransformatornotabschaltungsvorrichtung

“high-security transformer emergency shutdown device.”

 

[MattKing]

The translation functions - whether AI based or not - have been great for Photrio. My sense is that they permit a lot of non-English and limited English speakers to participate here.
As for idioms, English is so varied that sometimes it seems necessary to translate the posts of even the English speaking members

 

[Andreas Thaler]

The translation functions - whether AI based or not - have been great for Photrio. My sense is that they permit a lot of non-English and limited English speakers to participate here.

I had a bad experience with this on Reddit. Combined with my preference for well-structured text with bullet points, the most common comment was “AI slop.” But maybe they were actually referring to the content itself. But no further explanation was given.

As for idioms, English is so varied that sometimes it seems necessary to translate the posts of even the English speaking members

Until recently, I thought "the next fucking level" was a vulgar expression, until someone set me straight

 

[Ian Leake]

Anyway I can’t hide the fact that I’m a German (Austrian) native speaker, but here on the forum I don’t use the compound words that are popular in German, such as

Hochsicherheitstransformatornotabschaltungsvorrichtung

“high-security transformer emergency shutdown device.”

I, like many English speakers, suffer from hippopotomonstrosesquippedaliophobia

 

[Andreas Thaler]

Eigentlich würde ich ja viel lieber auf Deutsch schreiben. Aber für meine Reparaturberichte kenne ich die Übersetzungen für die englischen Fachbegriffe ins Deutsche meist nicht, die ich dem SPT Journal und den Reparaturhandbüchern der Hersteller entnehme.

Es ist nicht einfach

---

Actually, I’d much rather write in German. But when it comes to my repair reports, I usually don’t know the German translations for the English technical terms I take from the SPT Journal and the manufacturers’ repair manuals.

It’s not easy

 

[Andreas Thaler]

I, like many English speakers, suffer from hippopotomonstrosesquippedaliophobia

As a Swiss citizen, you have German, Italian, French, and Rhaeto-Romanic at your disposal.

What a rich variety of languages!

 

[4season]

Until recently, I thought "the next fucking level" was a vulgar expression, until someone set me straight

English is actually pretty simple. For example, a single piece of fruit is "fruit". Multiples are "fruits", except when they're not.

 

[Andreas Thaler]

English is actually pretty simple. For example, a single piece of fruit is "fruit". Multiples are "fruits", except when they're not.

Die Frucht - die Früchte

German is more complicated

 

[Andreas Thaler]

After asking Google Gemini a few questions about electronic SLR circuits and general topics related to camera repair, I’ve come to the following conclusion:

• helpful for searching through existing documents,
• generally useful for checking existing texts for factual accuracy, plausibility, and clarity,
• quite reliable for querying documented fields of knowledge such as electrical engineering,
• a source of creative input for problem-solving,
• but requires critical oversight, citation of sources, and independent verification in all cases.

All in all, it’s another useful tool in my toolbox that I have to manage and take responsibility for myself. But it’s by far no magic wand that solves all problems on its own.

It’s nice to still have a role to play at work