Am I soon to be redundant as editor of Macfilos? Will this blog be written soon by a bot, bringing together the cumulative experience of a million photographers instead of just one? If ChatGPT is anything to go by, the writing is on the wall. Entertaining as this AI is, though, I can’t help feeling that it will be a few years yet before the takeover.
In his recent Macfilos article on the use of artificial intelligence in creating photographs, Keith James asked — tongue in cheek, of course — whether photography is redundant. Can we hallucinate our photographs in future? Now, the question is whether or not we can hallucinate a blog such as Macfilos, technical references and all.
According to The Times, over a million users have tried out ChatGPT since its launch on November 30. One of them is your editor, experimenting with some pretty esoteric stuff on behalf of Macfilos readers.
While dreaming up his pictures with the aid of his DALL-E2 pal, Keith was driven to use “amazing” to describe some of the results. Similarly, I was pretty amazed by the stuff that ChatGPT presented in response to some rather difficult Leica-linked questions.
The output is not without its misinterpretations and downright factual errors, but bear in mind this is very early days. What will this three-week-old infant be capable of when it reaches maturity? Will it be writing novels, taking over convincingly from journalists, or pulling the wool over our eyes in a big way?
Having seen these little baby steps, I wouldn’t be at all surprised. It’s learning all the time.
ChatGPT has been developed by OpenAI, a San Francisco company founded by Elon Musk. Fortunately, the chatbot has an ethical filter, so it refuses to answer questions on illegal activity or evil intent. And at first glance, it is more useful than Tesla’s “full self-driving” package.
So here goes. Laugh at the silly errors, but don’t dwell on them. Concentrate, instead, on what this chatbot does get right. It is indeed pretty impressive.
Start by asking ChatGPT to write about a subject or to answer a question. It’s a bit like interrogating Google but far more interesting. Let’s go for it…
Chat GPT: Write a short review of the Leica Q2 Monochrom
“The Leica Q2 Monochrom is a high-end compact camera designed for photographers who value image quality and craftsmanship. It features a full-frame monochromatic sensor, which means it captures images in black and white only.
“The sensor has a resolution of 47.3 megapixels and is paired with a fixed 28mm f/1.7 lens, which is equivalent to a 35mm lens on a full-frame camera. The lens has a fast aperture, making it suitable for low-light photography and shallow depth-of-field effects.
“The camera also has a number of advanced features to enhance image quality and make it easier to use. It has a high-resolution electronic viewfinder, a touchscreen display that can be tilted, and built-in Wi-Fi for transferring images to a smartphone or tablet. It also has a weather-sealed body and is built to withstand the rigours of professional use.
“Overall, the Leica Q2 Monochrom is a well-regarded camera that offers excellent image quality and build quality. It is particularly suitable for photographers who are interested in black-and-white photography and want a compact camera with a fixed lens. It is a relatively expensive camera, so it may not be the best choice for photographers on a budget.”
ChatGPT: How does the Leica Q2 Monochrom compare with the Leica Q2
“The Leica Q2 Monochrom is specifically designed to capture high-quality black-and-white images, whereas the Leica Q2 is a colour camera. The Monochrom version has a monochromatic sensor, which means it captures images in black and white only. This allows it to capture a greater dynamic range and produce images with more detail and contrast compared to a colour camera that captures the same scene.
“The Monochrom version also has a higher resolution sensor compared to the Q2, with a resolution of 47.3 megapixels compared to the Q2’s 47 megapixels. This can result in slightly more detailed images, although the difference may not be noticeable in most cases.
“In terms of other features, the Q2 Monochrom and Q2 are very similar. Both have a fixed 28mm f/1.7 lens, a high-resolution electronic viewfinder, a touchscreen display, and built-in Wi-Fi. They also have a similar design and build quality.
“Overall, the Q2 Monochrom may be a better choice for photographers who are primarily interested in black and white photography and want a camera that is optimised for this type of shooting. The Q2 may be a better choice for photographers who want to be able to capture both colour and black and white images.”
So far, so good, barring the howlers. Let’s try something a little more esoteric:
Chat GPT: Write about Ernst Leitz and the way he helped Jewish employees in World War II
“Ernst Leitz was a German businessman and industrialist who founded the camera and microscope company Ernst Leitz GmbH (now known as Leica). Leitz was born in 1843 in Wetzlar, a small town in central Germany known for its precision optics industry.
“In the 1930s, as the Nazi party rose to power in Germany, Leitz was faced with a difficult dilemma: how to protect his company and its employees, many of whom were Jewish, from the increasingly oppressive policies of the Nazi regime. Leitz was an avowed opponent of the Nazis and their policies, and he made a conscious decision to protect his Jewish employees from persecution.
“Leitz implemented a number of measures to protect his Jewish employees, including transferring Jewish workers to other divisions of the company and giving them “non-essential” job titles to make them less likely to be targeted by the Nazis. He also used his influence and connections within the Nazi party to lobby for the protection of his Jewish employees.
“Despite these efforts, Leitz was unable to fully protect his Jewish employees from the consequences of Nazi persecution. Many of them were eventually forced to flee the country or were arrested and sent to concentration camps. However, Leitz’s efforts to protect his Jewish employees were seen as exceptional at the time, and his actions have been celebrated as a rare example of moral courage and resistance during the Nazi era.”
Glaring errors
As I mentioned above, ChatGPT does get the wrong end of the stick. There are glaring errors in the description of the Q2 and Q2 Monochrom. The suggestion that one has a larger sensor of 47.3MP compared with 47MP (two different sets of reference material pointing to the same sensor) is used as the basis of a spurious argument. And the reference to the 28mm lens being equivalent to 35mm “on a full-frame camera” is nonsense; where it got this idea from is anyone’s guess. The Q2 gets a fictitious flip screen. Overall, though, it’s a pretty good effort.
The Leitz piece is even more impressive, especially considering the niche subject matter. Apart from imagining that the original Ernst Leitz was still running the company at the age of over 100, ChatGPT is pretty factual, if a little bland.
Now imagine I decide to do a quick article on the history of the Leica M system and the benefits of using a rangefinder camera. Is this where ChatGPT goes in like a bull in a china shop?
ChatGPT: Review the history of the Leica M manual focus lenses
“Leica M-mount lenses are a series of interchangeable camera lenses designed for use with Leica’s M-series rangefinder cameras. These lenses are known for their high quality and compact size, and they have a long and storied history dating back to the 1950s.
“The first Leica M-mount lens was introduced in 1954 and was called the Summarit. This lens was a 50mm f/2.8 design and was followed by several other focal lengths, including 35mm, 90mm, and 135mm lenses.
“In the 1960s, Leica released a number of new M-mount lenses, including the Summicron, which featured a faster maximum aperture and improved image quality. This was followed by the release of the Summilux, which was an even faster lens with a maximum aperture of f/1.4.
“In the 1970s, Leica introduced many new M-mount lenses, including the Elmarit, which was a lighter and more compact lens, and the Tele-Elmarit, which was a telephoto lens with a maximum aperture of f/2.8.
“In the 1980s, Leica released the R-series of lenses, which were designed for use with their R-series SLR cameras. These lenses were not compatible with M-series cameras, but they did feature electronic aperture control and other advanced features.
“In the 1990s and 2000s, Leica continued to release new M-mount lenses, including the Summarit-M, which was a more affordable alternative to the Summicron, and the Noctilux, which was a super-fast lens with a maximum aperture of f/0.95.
“Today, Leica offers a wide range of M-mount lenses for its rangefinder cameras, including wide-angle, standard, and telephoto lenses, as well as speciality lenses for various applications. These lenses are known for their high build quality, excellent image quality, and compact size. They continue to be popular among photographers who value the tactile experience of manual focus photography.”
Well, poor old ChatGPT gets its neurons in a bit of a twist with this one. There is a clear obsession with the Elmarit and some misleading information. I’ll give this a three out of ten.
But maybe the ever-ready bot will do a bit better with this question (one which has been occupying Macfilos comments columns of late):
ChatGPT: Explain why rangefinder focus is superior to autofocus
Rangefinder focusing is a method of focusing a camera in which the distance to the subject is measured by a separate rangefinder, which is then used to set the focus of the lens. This method is often considered superior to autofocus (AF) for a number of reasons:
- Speed: Rangefinder focusing is generally faster than AF, particularly in low light or low contrast situations. This is because the rangefinder does not rely on the camera’s AF system to measure the distance to the subject, which can be slow and unreliable in these conditions.
- Precision: Rangefinder focusing allows the photographer to be very precise in their focusing, as they can see the distance to the subject in the viewfinder and adjust the focus accordingly. This can be particularly useful for precise focusing on small or distant subjects.
- Manual control: Rangefinder focusing allows the photographer to manually control the focus of the lens, which can be useful in situations where the AF system may not be able to focus on the intended subject accurately.
- Continuous focus: Some rangefinder cameras have the ability to continuously focus on a moving subject, which can be useful for action or sports photography.
- Overall, rangefinder focusing can be a good choice for photographers who need fast, precise, and manual control over the focus of their camera. However, it is worth noting that rangefinder cameras are generally more expensive and less common than AF cameras and may not be suitable for all types of photography.”
I think this is one for David Babsky to comment on, especially on the subject of continuous focus for moving subjects.
What does all this prove? Well, as we have discovered, Editor Evans isn’t yet redundant. Yet. Give him a year or two, and ChatGPT can take over. But it is astonishing that this chatbot can produce some convincing stuff after less than a month in the public domain. I imagine this is especially so if you ask it to write about more general subjects. Try asking it to write a romantic story about two of your friends, for instance, and then stand by for blasting. My questions must be pretty unusual, but I think ChatGPT has made a good fist in answering.
It will also not be lost on you that you can obtain similar essays on static websites such as Wikipedia. But the attraction of ChatGPT and other such systems as they emerge is that the information is gathered from many sources into a unique custom-tailored story. The writing style is factual (even if wrong) but rather lacking in flair. It isn’t anyone’s copyright; it can be adjusted and amended until it becomes an entertaining piece of writing.
Even now, I can see the use of the system in planning Macfilos articles. I am toying with the idea of asking ChatGPT for a summary before writing. Much pertinent reference material is produced, thus cutting down the research time. But, as with language translation bots, it helps to know what you are doing and have the capability to recognise egregious errors. When it comes to the complexities of the Leica M System, however, ChatGPT steps in where angels fear to tread.
Are you ready for ChatGPT yet? All you have to do is register and start asking questions. I warn you, though, this is another of those rabbit holes down which you will disappear in a trice…
ChatGPT: How do you disappear down a rabbit hole?
“To disappear down a rabbit hole means to become deeply engrossed or absorbed in something, often to the point of losing track of time or becoming isolated from the outside world. This phrase is often used to describe the feeling of being drawn into a complex or engaging activity or topic, such as reading a book, watching a movie, playing a game, or exploring an interesting website. It can also refer to the feeling of being caught up in a situation that is confusing or difficult to understand or that requires a lot of time and effort to resolve.”
Quite. Beam me up, Scotty…
What do you think? While neither DALLE-2 nor ChatGPT are ready for prime time, they point convincingly to the future. The developers quickly point out that both systems are in their infancy, but by releasing them for a sort of public beta-testing, they hope that the AI will learn and gradually iron out the obvious problems.
Want to contribute an article to Macfilos? It’s easy. Just click the “Write for Us” button. We’ll help with the writing and guide you through the process.
As a retired Un9versity professor, this article in the Guardian interested me:
https://www.theguardian.com/technology/2022/dec/31/ai-assisted-plagiarism-chatgpt-bot-says-it-has-an-answer-for-that
”
In a lecture at the University of Texas, OpenAI guest researcher Scott Aaronson said that the company was working on a system for countering cheating by “statistically watermarking the outputs”. The technology would work by subtly tweaking the specific choice of words selected by ChatGPT, Aaronson said, in a way that wouldn’t be noticeable to a reader, but would be statistically predictable to anyone looking for signs of machine-generated text.
“We want it to be much harder to take a GPT output and pass it off as if it came from a human,” Aaronson said. “This could be helpful for preventing academic plagiarism, obviously, but also, for example, mass generation of propaganda – you know, spamming every blog with seemingly on-topic comments supporting Russia’s invasion of Ukraine without even a building full of trolls in Moscow. Or impersonating someone’s writing style in order to incriminate them.
“We actually have a working prototype of the watermarking scheme,” Aaronson added. “It seems to work pretty well – empirically, a few hundred [words] seem to be enough to get a reasonable signal that, yes, this text came from GPT.” “
Does this do copyright violations?
Or to put it another way, can it detect when the use of material might be a breach of copyright? Increasingly, it is difficult to maintain copyright on anything that is on the internet. Earlier this year I bought an expensive book on brass lenses which contained a photograph of mine which had appeared in a Macfilos article. It is attributed to me, but my permission to use it had not been sought. The book also contained references to material, in articles I had written for Macfilos, for which permission to use was similarly not sought.
To put it another way, it is very difficult to find original material these days. As a result of the existence of IT and the Internet there is a lot of copy and paste going around, which is basically what this ChatGPT thing does.
William
We’ve had a very different approach to ChatGPT in that we gave it sensible parameters and asked it to write functions to achieve various tasks . . . in various different languages (Cobol, Visual Foxpro, PHP etc).
It wrote some extremely good code, and was open to lots of suggestions and modifications. It was also scarily good at debugging code which one gave to it.
So that whilst I agree with William and Stephen, Brian and David B that it’s not yet a great journalist, it actually works really well when given an empirical task.
I think that I’d be really quite worried if I was a jobbing programmer, and I wonder about the millions of programmers in India and other third world countries.
best
Jono
My programming ..or code-writing.. years are long gone, but I remember around 1980 or ’81 there was a popular (? ..but how popular?) program called ‘The Last One’, touted as the last program you’d ever need, as it could write whatever other program you ever wanted! ..It disappeared about two or three years later, as new programming languages overtook it, with the introduction of APL, B, BCPL, C, C+, C++ ..and at which point I also gave up writing code!
If it’s great – sorry; stopped there to unwrap presents for a while – if it’s great at writing code ..marvellous ..until it takes us to Skynet ..and you know what happens then: the machines take over and Arnie comes back from the future to kill all human resistance! ..Or we’re all plugged into cocoons to power the computers, and must choose the red pill or the blue pill to fix our future (..sorry; carried way into film-nerd mode there).
But without being able to understand human experience, rather than just read what people say about it, would ChatGPT-alike software, or neural nets, ever know what they’re writing documents about?
Only the other month, you’ll remember, Blake Lemoine was sacked from Google’s AI project LaMDA (language model for dialogue applications) chatbot research, as he believed that it was actually sentient ..not understanding, himself, that it was just regurgitating absorbed info back to him – rather like the original Eliza ( still available at tinyurl.com/3l1zab0t ) – but Blake, who’d been programming it, was thoroughly taken in by the chatbot’s ability to ..well, to chat!
What is the value of machine-written code, if it has no understanding of human experience, or human overseers have forgotten to include human experiences and forgetfulness?
Self-driving code is great, unless it’s been forgotten to include the possibility of cyclists losing their balance and wobbling out into the path of cars ..the code needs to know that what looks like a cyclist ahead may suddenly pull out, or may overtake on the left while the car’s approaching a left turn, or that the dog in the cyclist’s handlebar-basket may suddenly leap out, or the driver behind, fed up with the speed of the self-driving car, may pull out, overtake, and then pull in directly in front of the self-driven car to ‘teach it a lesson’.
Human frailties may need accommodating in computer-program-written code.
Now off for some Christmas pudding.
Where do I begin? In my experience AI is only ‘artificial’ and not ‘intelligent’ in the human sense. We have seen this with Google for many years. It started off as a search engine, but now it is a marketing tool which assumes that every search has a purchase as its end point. The examples above are joyless as they are just repetition of stuff already assembled by humans, but comes out without any editing or refinement or, indeed, discernment. Kathy’s experience with the tablet is a classic. Not only is there no discernment, but there are arrogant assumptions made by the ‘Fruity Ones’ working for the ‘Big A’. In moving forward with AI we need to concentrate on the ‘I’ aspect before we let it take over our lives. After all do we want a device that will not only devise and make our breakfast, but eat it as well?
Could I say that I suspect that most people here would not welcome such a joyless and soulless future run by AI? As regards the RF v AF discussion I was at pains to point out to David that my view was ‘personal opinion’, something you will not find with AI.
William
It’s only as good as its sources, and software updates are the bane of this kind of technology.The reason to enjoy reading articles is because of the experience and unique writing style the author employs. Why do I want to read AI reviews? No thanks. Last week, my Epson printer suddenly decided to use AI ( Absence of intelligence ) to determine that the paper tray was not loaded correctly ( It was ) resulting in me wasting more than one hour trying to defeat the so-called intelligence of the machine. Tech has a habit of doing one thing you want it to do and half a dozen things you don’t want it to do. Makes loading a roll of film in a ‘new’ M6 a delight.
With the quality of real journalism going down the toilet over recent years, it looks like most of the media can be easily replaced by this “tool”.
I wish all a very Merry Christmas except for Leica who said Merry Holidays in their release. I sent them an appropriate reply. Leica marketing will get nothing right. Luckily Leica builds brilliant products.
A very Merry Christmas and very blessed 2023 to Mike. Please take a few days off to recharge those grey cells as Poirot would say.
Speaking of . . . it’s a great deal of fun to watch a Poirot episode on YouTube, and turn on automated subtitles. But, everyone has to take a drink when ‘Poirot’ is (mis)translated. One can get quite tipsy ..
Happy Holidays to all, with dreams of new Leica products dancing in one’s head.
You asked for it, Mike..
“..this chatbot can produce some convincing stuff after less than a month in the public domain..”
Yup ..in the public domain, where it ‘scrapes’ material from every press release, blog, opinion piece, quote and misquote which it can find. But, of course, it has never actually itself handled a camera, and so it doesn’t really know what are, for example, the differences between using a rangefinder, using an SLR and using a toffee hammer.
If it can’t find a description of, say, the visible out-of-focus blur which a human can see on a built-in matte glass, or plastic, screen inside an SLR’s finder (..where you see exactly just how blurred any objects at out-of-focus distances may be..) then it can just take at face value any misleading comments which it finds about, for example, a rangefinder being a better camera for achieving o-o-f separation between different elements of a photo.
It’s only as good – or bad! – as the material which it finds. And if it doesn’t find accurate material, then it regurgitates inaccurate so-called info. Like the ‘echo chambers’ of (anti-)social media. If it reads that – as Putin says – the United States are responsible for the fighting in Ukraine ..then it may simply regurgitate that ..as fact! It has no personal experience of a physical world ..it has no objective experience of anything. As the old expression has it, from the earliest days of computers: GIGO – Garbage In, Garbage Out.
The trouble is, those who don’t know that all it produces is just a mashup of supposed info, and those who’ll believe anything that they read, will swear that it’s giving ‘real’ information. Read Wikipedia, and it’ll tell you as supposed ‘fact’, that I closed my cinema because a new multiplex was opening nearby. But the truth – which Wikipedia doesn’t display unless there’s already a published source which mentions it – is that my parents in Manchester both died, and so I sold it, and went back home to my Beloved in London.
Any bot which quotes only some story which has already been published, doesn’t, of course, know about things which have NOT been published. So if ChatGPT writes material which is then used as a source by, say, Wikipedia, then ChatGPT’s – and similar software’s – inaccurate misinformation may easily be presented as if from a reliable source.
Drivel then becomes mainstream accepted wisdom. Look at the States and Chump. Consider the reams of Brexit disinformation. Doctored photos are thought to be ‘true’. Lies are ‘accurate reporting’.
Look, there’s a nice door ..through which we enter the post-truth society.
Cheer up, folks ..the truth is now anything which you want to believe!
David,
Thanks for this overview. I think most of us would agree with the points you make. It is obvious that the system is regurgitating information that is available from many sources and, as you say, it doesn’t actually know what is right and what is wrong.
It’s rather like using AI translations, which actually work quite well these days. Translating into your own language is relatively painless simply because you know what is right and what is wrong (literal translation of idiom and proverbs, for instance, is very difficult). But to translate INTO a foreign language is a minefield unless you know enough of that language to spot the howlers. Still, it works after a fashion as long as you keep things simple, short sentences and avoid difficult constructions.
Chat GPT is at the moment a curiosity and something that should be used sparingly, especially if you know nothing about the subject. I think the examples in the article underline this. We spot the errors, but someone who knows nothing about Leica or photography would think it all perfectly acceptable.
Hi Mike, it’s a relief to know that ‘the powers that be’ at Macfilos will not be ushering you off stage left, anytime soon. Wait, I suppose you are the powers that be at Macfilos… Nevertheless, it is reassuring that you will still be at the helm for the foreseeable future, at least until ChatGPT 2.0 arrives.
It really is great fun playing with this app. It can provide coherent responses to queries that far out do Google and Wikipedia. I asked it why people still buy Leica cameras, even though they are very expensive. Its response was impressive, and made me feel much better about the amount of money I have spent on my Leica gear!
Merry Christmas and Happy Holidays to you and the Macfilos community! Cheers, Keith
Thanks for your confidence, Keith. As you gathered, I find all this quite fascinating and look forward to ChatGPT learning new tricks. One problem is that I don’t know how to tell the system that it is wrong — for instance on a factual error on the Q2/M. So, presumably, it perpetuates the error. There should be away of making corrections although, as with Wiki, that opens the way to misinformation and mischief.
I wish you all and all readers and contributors the very best for Christmas and the New Year.
Mike
Mike,
I’m less hopeful than you. I recently bought a ‘pro’ tablet from a well-known computer manufacturer (and very nice it is). It has a slot for a SIM card, which I don’t use; everything was fine until I upgraded to the latest operating system. Opening the tablet I got the kind of message we love from this manufacturer:
“Your SIM card is not installed. I am locking you out of your tablet until you acknowledge the error of your ways.”
It got tiresome, so I went to customer support, and explained I don’t have a SIM and don’t want one; that the problem was recent with the latest IOS (oh did I write that?) and what could I do?
I got back a very nice message telling me to make sure the SIM was fully inserted, etc, etc. Oh and make sure you have the latest IOS.
You can guess at what the chat-bot did: it picked out ‘SIM’, then attached all the text related to SIM cards. It ignored all my qualifiers.
What could I do? I replied “Arrigato, Mr. Robato.”
Kathy