alt Hacker News> A BBC journalist ran the image through an AI chatbot which identified key spots that may have been manipulated.
The image is likely AI generated in this case, but this does not seem like the best strategy for finding out if an image is AI generated.
Replies
Someone I know is a high school English teacher (being vague because I don’t want to cause them trouble or embarrassment). They told me they were asking ChatGPT to tell them whether their students’ creative writing assignments were AI-generated or not-I pointed out that LLMs such as ChatGPT have poor reliability at this; classifier models trained specifically for this task perform somewhat better, yet also have their limitations. In any event, if the student has access to whatever model the teacher is using to test for AI-generation (or even comparable models), they can always respond adversarially by tinkering with an AI-generated story until it is no longer classified as AI-generated
ChatGPT: This looks like AI. I can tell from some of the pixels and from seeing quite a bit of training data in my time.
This is the fast way they can try, but it shouldn't be the most trustworthy way and shouldn't be in report.
If it's nano banana you can give it to Gemini bc it has artifacts
Yeah, talk about begging the question. Yikes.
Yeah that hardly talks of the "journalist" being good at their job. At worst they asked a biased question like "has this photo been AI generated and if then how" or worse.
People tend to think that AI is like a specific kind of human which knows other AI things better. But we should expect better from people that do writing as their job.
Yeah, it is frankly just plain bad epistemology to expect an AI chatbot to have answers on a matter such as this. Like trying to get this week's lotto numbers by seeking a reading in bible passages and verses. There is no way that the information was encoded within in there as it would violate causality. At best you'd have coincidental collisions only.
Do you not think even BBC "journalists" are suffering from immense pressures to use AI for efficiency? It's everywhere
Under the other photos it says A photo taken by a BBC North West Tonight reporter showed the bridge is undamaged and A BBC North West reporter visited the bridge today and confirmed it was undamaged
They may have first ran the photo through an AI, but they also went out to verify. Or ran it after verification to understand it better, maybe