When I asked gemini to randomly arrange the numbers between 4 and 27, it spit out a seemingly correct list of numbers with the issue that 23 was randomly missing
Jupp, as ai is trained on humans it will inherit our biases. That’s one of the biggest problems to solve. “If we train our ai on 4chan posts, how do we make it not racist\sexist\etc.”
LLM based technology has been shown to have biases in randomness, there was an article a while back experimenting with coin flips showing a lack of true randomness.
Its cause it’s about token prediction, so their is forced priority behind the scenes whether or not that is visible to the user.
Its the same reason why when you ask an image generator to create “a person from India” you get a man in a turbin a majority of the time.
A lot of 3s and no 7. Does AI has a bias on what numbers they create?
Like if I generate 1000 pictures with a number between 0 and 9, are those numbers distributed equally or what would the distribution look like?
Humans, when ask to say random numbers also have biases in some circumstances, so I guess AI does too.
When I asked gemini to randomly arrange the numbers between 4 and 27, it spit out a seemingly correct list of numbers with the issue that 23 was randomly missing
Jupp, as ai is trained on humans it will inherit our biases. That’s one of the biggest problems to solve. “If we train our ai on 4chan posts, how do we make it not racist\sexist\etc.”
LLM based technology has been shown to have biases in randomness, there was an article a while back experimenting with coin flips showing a lack of true randomness.
Its cause it’s about token prediction, so their is forced priority behind the scenes whether or not that is visible to the user.
Its the same reason why when you ask an image generator to create “a person from India” you get a man in a turbin a majority of the time.
4 also appears 3 times, but that number 3 isn’t always a number 3 - especially the bottom right kinda looks like a negative 3, and the one left of it…