This was like AI before AI. How does this french magic work?
For me it is ALICE
It just has records of who answered the same to questions and if it didn't get it right you eventually tell it. Then someone else goes through it and leads to that answer. Hardly intelligent.
>>109240152so it's just fable without the jewery, and extra steps, got it
>>109240152Oh really? That’s disappointing
>>109240125>How does this french magic workif ()else {}
The answer to each question progressively rules out large portions of all of the entries in it's huge list. After a surprisingly low number of questions you can, in most cases, narrow down to a very small list of candidates.If your answer is something that's not on the list yet, it's placed onto the list and tagged with the 'side' of each question you answered during the game.After doing this for years and millions of game runs, it looks like magic. But it's not.
>>109240302it's more likely some kind of scoring system.I don't think a non-matching answer completely excludes a character.
nowadays it just uses neuron networks like everybody else, but it lets itself being trained in real time by its users instead of working on a pre trained model like popular chatbots
>>109240288It's a really simple tree algorithm. It just relies on millions of users adding thousands of branches by giving it the correct answers.You unknowingly do the same thing with Guess Who.
>>109240716no, it doesn't. why do you just make shit up?
>>109240125It's a secret project joining AWS, Facebook, and Google's decades of tracking your actions over the years to predict your thoughts.
it's a scoring system where the question it asks you is the one that has the largest chance of decreasing the set of possible results, or more accurately, to decrease the amount of close matches vs distant matches
>>109240125Holy shit, Akinator is Fr*nch? Jesus Christ
>>109240125Take a simple question, like if the person is male/female. That's going to reduce the possibilities by ~50%. If they're real or imagined, also rules out a large amount. With enough of these types of questions, each time you're narrowing down the possibilities by a huge amount. Once you get to 20, it doesn't seem like a lot but it actually really narrows it down more than you might think.You know like the folding a piece of paper analogy, if you could actually fold it enough times like 50 times, and it's doubling in height each time then it's enough to reach the moon or something like this. There is a similar principle here at work with the questions,I don't know if they actually hand crafted a dataset, I wouldn't be surprised if they actually had people play the game to create the dataset itself, i.e. if you picked some say fictional superhero or celebrities, and ran them through the questions and it "did I get it right?" is actually helping to add to the dataset, i.e. it learns that Spiderman is fictional, male, etc whereas Scarlett Johansen is real, female, etc. Purely from working backwards from the data its users got. Another interesting thing is that it once it gets "good enough" it can go viral because people share it with others.Would this approach of crowd sourced data work today? Probably not. If they were building a dataset based on working back from the questions, then you're going to have non-whites lying at the end to fuck with it.
>>109240125Lots and lots and lots of data.
you literally tell it what your character is.
it had CP "actresses" in there at one time. can't be bothered to see if they wiped them yet.
