|
It's at four outta four for:
Savior
Tossgirl
Shinji (NGE)
Shana (SnS)
...will attempt more obscure anime.
|
Hahaha, it got kaepora gaebora from Zelda (the wise talking owl).
While we're at it, it got Zelda herself too ;o
Oh god, Mithos from Tales of symphonia! @_@
|
Regarding obscue anime: it got Dokuro Mitsukai (Bokusatsu Tenshi Dokuro-chan) right.
I was surprised until I read that in the last 7 days Yuki Nagato (from The Melancholy of Haruhi Suzumiya) was used 3305 times.
|
On January 06 2009 01:27 fanatacist wrote:Show nested quote +On January 06 2009 01:19 Klive5ive wrote:On January 06 2009 01:07 fanatacist wrote:On January 05 2009 20:41 Kwark wrote:On January 05 2009 16:42 fanatacist wrote:On January 05 2009 16:40 Kuja900 wrote:
OK im bumping this because this thing is just amazing. He got tasteless correct and he didnt ask if he was even living in asia and barely anything related to him. The last question was can your character solve the rubics cube lol and he still somehow got tasteless right omg this thing is just insanity. Once he is relatively certain of th answer he starts to fuck with you to induce that kind of "wow" factor that makes it seem more magic than scientific from simpler audiences. That's not why it does that. If it went questions in a logical elimination way, ie male/female then real/non real it could do it much faster but it would never learn any new answers about the people it had. Every time you answer questions you add to its database. When it knows the answer it'll try and find out more about them. It's the random questions that help it differentiate between oov and nada. So you're telling me that whoever made this program didn't use conditionals on groups of questions? Example: If user says YES to nationality question, then NO is true for other nationalities. Also you're telling me that there are that many distinctions between oov and nada in his general list of questions? Not really, the majority of the times you try to get a progamer it guesses one at random because they share so many characteristics. Asking if Savior is French after it has already been determined that he is from Asia is not going to help the program pick him over oov. No clearly the program can't eliminate questions with other questions. I would assume this is because it doesn't have the negative answer to all it's questions. In other words it knows that Savior is Korean, so if someone says "The character is Korean" it puts savior as a higher possibility. However it doesn't know that John Cleese is NOT Korean so it DOESN'T eliminate him from the list. Therefore later on it might ask "Is your character English?" to eliminate him. The only way it could learn this would be to ask "Is your character Korean?" when the person is searching for John Cleese. People are born in only one nation so it would be logical and more productive to have a conditional on this. I would assume this because it is pretty unlikely that you would be the first person to say that Savior is NOT French, Of course it could be to double check its facts, however it seems much more logical that they would eliminate other options immediately after getting a correct answer and would just mess with you with questions it knows are wrong. If you look at his animation when he asks the question, much of the time it is already smug and self-confident when it asks you clearly wrong answers later in the game. I don't think this is coincidence.
it doesnt do that automatically. The program only know that certain questions imply certain characters more than others. It doesn't know the relationship between individual questions. In other words, it doesnt know that if you say "yes he's korean" the answer to "is he english?" is automatically known.
|
On January 06 2009 04:39 fusionsdf wrote:Show nested quote +On January 06 2009 01:27 fanatacist wrote:On January 06 2009 01:19 Klive5ive wrote:On January 06 2009 01:07 fanatacist wrote:On January 05 2009 20:41 Kwark wrote:On January 05 2009 16:42 fanatacist wrote:On January 05 2009 16:40 Kuja900 wrote:
OK im bumping this because this thing is just amazing. He got tasteless correct and he didnt ask if he was even living in asia and barely anything related to him. The last question was can your character solve the rubics cube lol and he still somehow got tasteless right omg this thing is just insanity. Once he is relatively certain of th answer he starts to fuck with you to induce that kind of "wow" factor that makes it seem more magic than scientific from simpler audiences. That's not why it does that. If it went questions in a logical elimination way, ie male/female then real/non real it could do it much faster but it would never learn any new answers about the people it had. Every time you answer questions you add to its database. When it knows the answer it'll try and find out more about them. It's the random questions that help it differentiate between oov and nada. So you're telling me that whoever made this program didn't use conditionals on groups of questions? Example: If user says YES to nationality question, then NO is true for other nationalities. Also you're telling me that there are that many distinctions between oov and nada in his general list of questions? Not really, the majority of the times you try to get a progamer it guesses one at random because they share so many characteristics. Asking if Savior is French after it has already been determined that he is from Asia is not going to help the program pick him over oov. No clearly the program can't eliminate questions with other questions. I would assume this is because it doesn't have the negative answer to all it's questions. In other words it knows that Savior is Korean, so if someone says "The character is Korean" it puts savior as a higher possibility. However it doesn't know that John Cleese is NOT Korean so it DOESN'T eliminate him from the list. Therefore later on it might ask "Is your character English?" to eliminate him. The only way it could learn this would be to ask "Is your character Korean?" when the person is searching for John Cleese. People are born in only one nation so it would be logical and more productive to have a conditional on this. I would assume this because it is pretty unlikely that you would be the first person to say that Savior is NOT French, Of course it could be to double check its facts, however it seems much more logical that they would eliminate other options immediately after getting a correct answer and would just mess with you with questions it knows are wrong. If you look at his animation when he asks the question, much of the time it is already smug and self-confident when it asks you clearly wrong answers later in the game. I don't think this is coincidence. it doesnt do that automatically. The program only know that certain questions imply certain characters more than others. It doesn't know the relationship between individual questions. In other words, it doesnt know that if you say "yes he's korean" the answer to "is he english?" is automatically known.
I'm saying that it would make sense for it to do that, because I imagine the questions are grouped in the programming to randomly select X nationality out of Y list for question, then if X = 1 (true) then Y-X = 0 (false). This would definitely be the best way to actually narrow it down.
|
On January 06 2009 05:07 fanatacist wrote:Show nested quote +On January 06 2009 04:39 fusionsdf wrote:On January 06 2009 01:27 fanatacist wrote:On January 06 2009 01:19 Klive5ive wrote:On January 06 2009 01:07 fanatacist wrote:On January 05 2009 20:41 Kwark wrote:On January 05 2009 16:42 fanatacist wrote:On January 05 2009 16:40 Kuja900 wrote:
OK im bumping this because this thing is just amazing. He got tasteless correct and he didnt ask if he was even living in asia and barely anything related to him. The last question was can your character solve the rubics cube lol and he still somehow got tasteless right omg this thing is just insanity. Once he is relatively certain of th answer he starts to fuck with you to induce that kind of "wow" factor that makes it seem more magic than scientific from simpler audiences. That's not why it does that. If it went questions in a logical elimination way, ie male/female then real/non real it could do it much faster but it would never learn any new answers about the people it had. Every time you answer questions you add to its database. When it knows the answer it'll try and find out more about them. It's the random questions that help it differentiate between oov and nada. So you're telling me that whoever made this program didn't use conditionals on groups of questions? Example: If user says YES to nationality question, then NO is true for other nationalities. Also you're telling me that there are that many distinctions between oov and nada in his general list of questions? Not really, the majority of the times you try to get a progamer it guesses one at random because they share so many characteristics. Asking if Savior is French after it has already been determined that he is from Asia is not going to help the program pick him over oov. No clearly the program can't eliminate questions with other questions. I would assume this is because it doesn't have the negative answer to all it's questions. In other words it knows that Savior is Korean, so if someone says "The character is Korean" it puts savior as a higher possibility. However it doesn't know that John Cleese is NOT Korean so it DOESN'T eliminate him from the list. Therefore later on it might ask "Is your character English?" to eliminate him. The only way it could learn this would be to ask "Is your character Korean?" when the person is searching for John Cleese. People are born in only one nation so it would be logical and more productive to have a conditional on this. I would assume this because it is pretty unlikely that you would be the first person to say that Savior is NOT French, Of course it could be to double check its facts, however it seems much more logical that they would eliminate other options immediately after getting a correct answer and would just mess with you with questions it knows are wrong. If you look at his animation when he asks the question, much of the time it is already smug and self-confident when it asks you clearly wrong answers later in the game. I don't think this is coincidence. it doesnt do that automatically. The program only know that certain questions imply certain characters more than others. It doesn't know the relationship between individual questions. In other words, it doesnt know that if you say "yes he's korean" the answer to "is he english?" is automatically known. I'm saying that it would make sense for it to do that, because I imagine the questions are grouped in the programming to randomly select X nationality out of Y list for question, then if X = 1 (true) then Y-X = 0 (false). This would definitely be the best way to actually narrow it down.
yeah but who is going to form those groupings? A programmer would have to go in and tell the program, if someone is English, they aren't Japanese (this isnt always true, but even if it were the computer can't make that connection itself.) Which means that everytime one answer rules out another - is the character a girl would ideally rule out the character being a boy, someone has to go in and tell the program that.
When you have a bunch of user submitted entries, its just not realistic. It's impossible for a programmer to go through every dichotomy - "is your character in candyland?" - "then its not real", and hardcode which counters which.
The only reasonable way I could see it being accomplished is if the program noted that "99.99% of the time, people responded no to the question "is it real" and yes to "is it imaginary". It could then assume a dichotomy exists at the level of user queries it handles. It would still have to ask both questions occasionally though, incase the dichotomy is temporary and not permanent
|
machine learning at its best
|
On January 06 2009 05:28 fusionsdf wrote:Show nested quote +On January 06 2009 05:07 fanatacist wrote:On January 06 2009 04:39 fusionsdf wrote:On January 06 2009 01:27 fanatacist wrote:On January 06 2009 01:19 Klive5ive wrote:On January 06 2009 01:07 fanatacist wrote:On January 05 2009 20:41 Kwark wrote:On January 05 2009 16:42 fanatacist wrote:On January 05 2009 16:40 Kuja900 wrote:
OK im bumping this because this thing is just amazing. He got tasteless correct and he didnt ask if he was even living in asia and barely anything related to him. The last question was can your character solve the rubics cube lol and he still somehow got tasteless right omg this thing is just insanity. Once he is relatively certain of th answer he starts to fuck with you to induce that kind of "wow" factor that makes it seem more magic than scientific from simpler audiences. That's not why it does that. If it went questions in a logical elimination way, ie male/female then real/non real it could do it much faster but it would never learn any new answers about the people it had. Every time you answer questions you add to its database. When it knows the answer it'll try and find out more about them. It's the random questions that help it differentiate between oov and nada. So you're telling me that whoever made this program didn't use conditionals on groups of questions? Example: If user says YES to nationality question, then NO is true for other nationalities. Also you're telling me that there are that many distinctions between oov and nada in his general list of questions? Not really, the majority of the times you try to get a progamer it guesses one at random because they share so many characteristics. Asking if Savior is French after it has already been determined that he is from Asia is not going to help the program pick him over oov. No clearly the program can't eliminate questions with other questions. I would assume this is because it doesn't have the negative answer to all it's questions. In other words it knows that Savior is Korean, so if someone says "The character is Korean" it puts savior as a higher possibility. However it doesn't know that John Cleese is NOT Korean so it DOESN'T eliminate him from the list. Therefore later on it might ask "Is your character English?" to eliminate him. The only way it could learn this would be to ask "Is your character Korean?" when the person is searching for John Cleese. People are born in only one nation so it would be logical and more productive to have a conditional on this. I would assume this because it is pretty unlikely that you would be the first person to say that Savior is NOT French, Of course it could be to double check its facts, however it seems much more logical that they would eliminate other options immediately after getting a correct answer and would just mess with you with questions it knows are wrong. If you look at his animation when he asks the question, much of the time it is already smug and self-confident when it asks you clearly wrong answers later in the game. I don't think this is coincidence. it doesnt do that automatically. The program only know that certain questions imply certain characters more than others. It doesn't know the relationship between individual questions. In other words, it doesnt know that if you say "yes he's korean" the answer to "is he english?" is automatically known. I'm saying that it would make sense for it to do that, because I imagine the questions are grouped in the programming to randomly select X nationality out of Y list for question, then if X = 1 (true) then Y-X = 0 (false). This would definitely be the best way to actually narrow it down. yeah but who is going to form those groupings? A programmer would have to go in and tell the program, if someone is English, they aren't Japanese (this isnt always true, but even if it were the computer can't make that connection itself.) Which means that everytime one answer rules out another - is the character a girl would ideally rule out the character being a boy, someone has to go in and tell the program that. When you have a bunch of user submitted entries, its just not realistic. It's impossible for a programmer to go through every dichotomy - "is your character in candyland?" - "then its not real", and hardcode which counters which. The only reasonable way I could see it being accomplished is if the program noted that "99.99% of the time, people responded no to the question "is it real" and yes to "is it imaginary". It could then assume a dichotomy exists at the level of user queries it handles. It would still have to ask both questions occasionally though, incase the dichotomy is temporary and not permanent
I agree with what you're saying fusion.
Fanatacist you need to understand how algorithms work. It's not a case of saying "well it's more logical to do it like this". It might be more logical for a human but not a computer.
Dichotomy isn't quite the right word in relation to Countries of birth, but I know what fusion meant.
What we're saying is the program can't determine mutually exclusive characteristics.
A programmer could go through and create groups of mutually exclusive questions.
But each question would require another complicated attribute detailing what other questions are mutually exclusive. There are a number of problems with this:
1) At this point it would be almost impossible to alter the database.
2) The way the program works it appears to eliminate people and not questions. It then uses questions associated with the people left on it's list. Considering attributes associated with the questions would be complicated at this stage.
3) Checking for the attribute would take up processing time and slow down the program.
4) The programmers would have to keep constantly checking the incoming questions to see if they were mutually exclusive.
Not only this but eventually the program will learn the negative answer and stop asking the irrelevant questions anyway.
On a lighter note: It always asks me "is the character Brazilian" early on. I guess because I'm a male it assumes my character is a porn star or a football icon.
|
Lol, conditionals are NOT difficult in most practical languages, including ones with grouping exceptions.
|
yes, but who is going to write the groupings? The computer can't do it itself
there are almost a limitless amount of these groupings and it doesnt matter how easy each one is to code, for each and every single case where A being true implies B being untrue, the programmer has to go in and state that. Thats a lot of programming time wasted and a lot of code length.
When its user submitted questions/characters, there is no way for the computer to know that if something is an elf, its not a dwarf; if something is a vegetable it must be a plant; if something has long hair it must have hair; if its a girl then it must be female; if it sings then its a singer; if shes internationally famous then shes probably well-known....are you really going to hardcode those millions of lines into the program? and then update it every week to match the changing user submitted content?
the only reasonable way I can see is the sort of machine learning above, but that isn't perfect and has a number of drawbacks as explained above
some things are really easy for a human, and really difficult for a program. This is one of those things
|
Yeah the problem is the computer does not actually know the meaning of the questions it's asking. When it asks "Is the character female?", it doesn't actually know what it's asking. The computer does not know that being a female means you don't have a penis (probably), so it may very well later on also ask the question, "does this character have a penis?" .
Since the computer doesn't understand the meaning of the questions it asks, it can't group the questions based on that meaning. In that case, a human would have to manually program these groupings as questions are created. This is unreasonable because the point is that this program should be able to function automatically without human intervention. Also, if there are N questions, the space of possible groupings is 2^N. That is a massive amount of lookup that you would have to do.
One thing that the computer could do is prune out questions that are essentially identical or exactly opposite, based on the fact that the results from asking the question were always the same. For example, "Is this character male?" and "Is this character female?" should have nearly opposite results. From this, you can infer that these questions are opposites, and remove one of the questions from the question pool.
|
Holy shit its Savior. But I could've been thinking Boxer. Why did it guess Savior?
THIS IS MADNESS
|
Alright I would go further into this discussion but it's simply not worth any of our times. Until someone looks up the actual code we can speculate all day.
|
|
|
On January 06 2009 08:39 fanatacist wrote:
Alright I would go further into this discussion but it's simply not worth any of our times. Until someone looks up the actual code we can speculate all day.
continue if you want, I don't mind
how do you envision the code being written? How do you overcome the problems listed in the above post?
this entire thread you've been saying "this is how it should work" we respond with "It can't really work that way for these reasons:" you say "It will work" we say "the program can't innately know something like a human does" you say "unless you're the author this is all speculation"
It would be nice if you actually tried to resolve the issues we brought up. Otherwise it seems like you just want to get out of this topic without admitting you were incorrect.
|
Got Richard Rahl. Very nice.
|
I kept doing July and it kept guessing Bisu.
|
omg he guess Tossgirl in 20 question, without any linked too video games.
and he can't guess the singer Van Morrison with 50 questions ...
conclusion : TL.net own akinator
|
got Conan then Tintin then Kerringan then Iker Casillas
Amazing xD
|
i tried to do dorothy from wizard of oz and he hasnt gotten it!
edit: damnit tried it again and he got it!
|
|
|
|
|
|
EPT
