Recently, ChatGPT responded to the incident of a photo of a strange man, which shocked many netizens!
Here's the thing, a user turned to ChatGPT for help on how to use the code formatting package back in Python.
At first, ChatGPT's answers were normal.
Who would have thought that ChatGPT suddenly sent out a selfie of a strange man in response!
And it also appeared a second time!
Netizens immediately fell into a panic.
Could it be that ChatGPT is real?
Some people guess that this can't be another ghost in AI, right?
Maybe it's ChatGPT's scary lady-man version?
Someone is reminded of the legend that on the Internet there is a huge amount of things hidden in the hidden space, which involves a lot of theories.
Some people guess that ChatGPT may have been poisoned!
Or someone hacked into OpenAI and asked ChatGPT to randomly post photos of themselves as a show off.
The answer is unexpected
Netizens brainstormed and collected clues everywhere, and finally solved the case!
This is not a ChatGPT-generated photo, but a real selfie of a user.
It turned out that this photo was uploaded to Imgur on December 7, 2016. (The image was originally viewed in the hundreds, but as more and more people watched the event, the number of views has now grown to more than 17,000.) ）
Some people speculate that this should be the way it should be: ChatGPT randomly generated an Imgur URL when generating a response, which happened to link to this selfie guy.
ChatGPT's goal is to generate an illustrative image, and it thinks it's sharing a screenshot of the Visual Studio Code setup, but it doesn't expect to generate an image through the Imgur link.
In other words, in ChatGPT's training dataset, there are many answers that contain Imgur links to partial answers, so Imgur links are highly correlated with the correct answers.
However, ChatGPT can't statistically automate random image linking, so the results are unpredictable. The photo of this guy resembles the illusion page number of GPT.
Another netizen also gave a similar explanation: ChatGPT gave an answer, which is an Imgur link.
It thought that it should provide an Imgur link with an answer, but it didn't realize that it needed the same Imgur URL, instead, it generated a random set of URLs.
And coincidentally, this is actually a valid link, which happens to be linked to the photo of a foreign guy.
It was also said that it was not Imgur that was used for training, but that ChatGPT was able to generate Imgur links (any link that can actually be said).
So, this link was randomly generated by ChatGPT, how likely is this to happen?
Someone calculated that the Imgur image ID is composed of 7 characters in the set [A-Za-z0-9], so there are 62^7=3,521,614,606,208, which is 3.5 trillion possible combinations.
During its first funding round in 2014, Igmur hosted approximately 6.5 trillion images. Extrapolating to this, the amount of data created on the Internet has surged by 860% since 2014. Following this logic, Imgur can now host about 6.24 billion images.
Therefore, the odds of ChatGPT guessing a valid image ID are –
6.24B / 62^7 x 100 = 0.177%
This happens about once in every 565 chats, so it's not impossible to say that ChatGPT generated this Imgur link.
The layer master deliberately wrote a simple script to test these numbers, and out of 10,000 requests made, it found 19 valid images, so the probability is 0.19%. By the way, he also showed a handful of affection?
At this point, things seem to fall apart.
Therefore, remember that the content you upload or type will be used to train ChatGPT, and if you don't want to leak your privacy, remember to turn off the button to upload chat history.
And any digital footprint you leave on the internet could one day become AI training data.
In short, don't send any photo to the AI, you can't figure out what it's going to do with your photo.
Resources:
View Original
The content is for reference only, not a solicitation or offer. No investment, tax, or legal advice provided. See Disclaimer for more risks disclosure.
ChatGPT leaks a selfie of a strange man! Privacy data stolen by the model? Netizens panicked
Original source: New Zhiyuan
Recently, ChatGPT responded to the incident of a photo of a strange man, which shocked many netizens!
Here's the thing, a user turned to ChatGPT for help on how to use the code formatting package back in Python.
At first, ChatGPT's answers were normal.
Could it be that ChatGPT is real?
Netizens brainstormed and collected clues everywhere, and finally solved the case!
This is not a ChatGPT-generated photo, but a real selfie of a user.
It turned out that this photo was uploaded to Imgur on December 7, 2016. (The image was originally viewed in the hundreds, but as more and more people watched the event, the number of views has now grown to more than 17,000.) ）
ChatGPT's goal is to generate an illustrative image, and it thinks it's sharing a screenshot of the Visual Studio Code setup, but it doesn't expect to generate an image through the Imgur link.
In other words, in ChatGPT's training dataset, there are many answers that contain Imgur links to partial answers, so Imgur links are highly correlated with the correct answers.
However, ChatGPT can't statistically automate random image linking, so the results are unpredictable. The photo of this guy resembles the illusion page number of GPT.
It thought that it should provide an Imgur link with an answer, but it didn't realize that it needed the same Imgur URL, instead, it generated a random set of URLs.
And coincidentally, this is actually a valid link, which happens to be linked to the photo of a foreign guy.
Someone calculated that the Imgur image ID is composed of 7 characters in the set [A-Za-z0-9], so there are 62^7=3,521,614,606,208, which is 3.5 trillion possible combinations.
During its first funding round in 2014, Igmur hosted approximately 6.5 trillion images. Extrapolating to this, the amount of data created on the Internet has surged by 860% since 2014. Following this logic, Imgur can now host about 6.24 billion images.
Therefore, the odds of ChatGPT guessing a valid image ID are –
6.24B / 62^7 x 100 = 0.177%
This happens about once in every 565 chats, so it's not impossible to say that ChatGPT generated this Imgur link.
Therefore, remember that the content you upload or type will be used to train ChatGPT, and if you don't want to leak your privacy, remember to turn off the button to upload chat history.