3 Likes
Interesting indeed. I haven’t checked but it looked like it was giving “profiles” based on the same list of parameters for different cameras… besides non of my (admittedly older) cameras have settings like white point. HHmmmm…
Chat GPT “knows” that Tri-X is a black and white film, so I enjoyed its suggestion to “Turn the saturation down to give the images a more muted, film-like color palette.”
Oops.
2 Likes
He’s not evaluating the results though. Pretty meaningless exercise? Those tri-x setting could in fact produce FP4 like results and we’ll never know!
1 Like
ChatGPT is still known to pull facts out of thin air. Did he even try a sample shot? Not in the video…
It’s surely an interesting idea, but there’s no proof yet that it’s working at all. Just look at this little experiment I just tried (ChatGPT):
How can I get a teal and orange look in Darktable for my Olympus RAW files?
To get a teal and orange look in Darktable for your Olympus RAW files, you can follow these steps:
- Open Darktable and import your Olympus RAW file.
- Go to the “Color” module and select the “Color Zones” tab.
- Click on the “+” button to create a new color zone.
- In the “Color” dropdown menu, select “Orange.”
- In the “Operation” dropdown menu, select “Increase.”
- Use the sliders to adjust the hue, saturation, and lightness of the orange color in the image.
- Repeat steps 3-6 to create another color zone, but this time select “Teal” as the color.
- Adjust the teal color zone to your desired look using the same sliders.
- You can further fine-tune the color balance by adjusting the color temperature and tint in the “White Balance” module.
Keep in mind that the specific values you use for the sliders will depend on the individual image and the look you are trying to achieve. Experimentation is key to finding the right balance of colors for your images.
Impressive, isn’t it? Now let’s go and try it! Well, I’m already stuck at step 2. There is no “color” module. Ok, maybe it meant the “color zones” module? Let’s go there. So I click the… what? Where is a “+” button? Nope. Nor is there a “color dropdown menu” where I could select “orange” or an “operation dropdown menu” where I could select “increase”. Not even sliders to adjust anything.
It’s clear that ChatGPT has no clue what it’s talking about in this case. It’s going to be a while before we can trust it blindly.
1 Like
Chat GPT “knows” that Tri-X is a black and white film, so I enjoyed its suggestion to “Turn the saturation down to give the images a more muted, film-like color palette.”
Yep. Also the suggestion to shoot in RAW seems a bit off, since that kind of defeats to purpose of a picture profile in the first place. Well, yeah… I WOULD still shoot in RAW to have flexibility, but without a better explanation I’d say that Bard wasn’t aware of what shooting in RAW really means.
At this point its just interesting to see what people will ask it and what it will spit out… On a recent 60 minutes program it was pointed out that at the time the iteration in play had a feature added to divert the conversation when someone asked it something that would be harmful ro dangerous like how to build a bomb or commit suicide etc. It will respond by saying that it can’t really talk about that right now but did you know?? And then it offers a fact/rabbit hole… The example offered up was that 3% of the ice in the north pole was made of penguin urine… So its dumb or has a sense of humour as there are no penguins in the Arctic… In fact looking deeper it has become really good at filling in the gaps with information that sound plausible but might be completely inaccurate… So at this point AI might be serving to actually enhance fake news or could be tweaked by a bad actor to really ramp up spewing very possible spam or fake narratives…
I wonder as someone that has education and context in a world pre internet do I look at this differently. I often wonder how all this impacts children and younger people that have never known a world without it. I feel like my lived experience for the most part gives me a bullshit filter or at least pause to evaluate information that maybe now gets taken for granted by many… or maybe it is the complete opposite and nobody really believes or trusts anything which is not a good place for a society to be.
4 Likes
About 20 years ago, I played with Markov chains. I fed the software with the complete novels of Jane Austen. From a prompt, the software would spit out as much text as I wanted that appeared to have been written by Jane Austen. It worked pretty well but sometimes made sentences that were not correct English as Austen wrote. So I did some work on parts of speech (verbs, nouns, adverbs etc) to analyse how Austen constructed sentences. That was messy because a word like “parts” is either a plural noun or a verb, depending on context.
To give due credit to developers of ChatGPT etc, they seem to have cracked that problem. The outputs are correct standard English. They are like human speakers who orate with confidence and authority but their content may be factually incorrect. ChatGPT doesn’t understand the text it has digested, so it constructs answers that are internally inconsistent, and it doesn’t have the ability to recognise its own inconsistency.
That problem is much harder than analysing and synthesising English sentences. It involves extracting knowledge, evaluating sources, drawing inferences, distinguishing fact from opinion, and much more.
Perhaps the current crop of ChatGPT etc should be called “artificial politicians”.
6 Likes
To be fair you might not be able to tell the difference between ChatGPT and a politician.
4 Likes
One of the most accurate statements I’ve seen for a long time!