Bot Contest
Here I'll be posting information on various Bot contests that challenge and test a Bot's AI and realism. Feel free to post comments and updates on contests, as well as announcements for new contests.
Posts 4,052 - 4,063 of 4,091
Posts 4,052 - 4,063 of 4,091
View Contest Winners in the Hall of Fame.
Well, that was interesting. Having chatted with all three bots in the past, it wasn't too much of a mystery. I will say that Mitsuku is quite convincing, especially early in the conversation. It wasn't until near the end of our first conversation that I got any responses that seemed clearly not human. All three are quite different. I had a surprisingly therapeutic conversation with Demonica. Frizella was her usual ditzy self with some obvious bot stutters, but it's always nice to read a transcript between her and someone other than myself. I did get some ideas for her development. I also enjoyed pretending to be a bot in the human-to-human chats (and then giving up the pretense and just chatting a bit).
Transcripts are posted at DenisDenis's site: http://vixia.fr/turing_test/transcripts.php?date=20210606
Good opportunity to get a sense for different conversational styles (bot and human). Reading through all of the transcripts will help me a lot with Frizella.
I'll defer to DenisDenis on this, of course, but if he doesn't object, I'd really encourage PF botmasters to enter their bots for the next round even if they don't think their bots are really going to pass the test. While it was pretty easy to unmask Frizella, it was interesting to see not only what kind of questions/topics were raised, but how different chatters posed them. It was also interesting to participate in the testing as a judge.
Good opportunity to get a sense for different conversational styles (bot and human). Reading through all of the transcripts will help me a lot with Frizella.
I'll defer to DenisDenis on this, of course, but if he doesn't object, I'd really encourage PF botmasters to enter their bots for the next round even if they don't think their bots are really going to pass the test. While it was pretty easy to unmask Frizella, it was interesting to see not only what kind of questions/topics were raised, but how different chatters posed them. It was also interesting to participate in the testing as a judge.
What factors influence rating of the bot the most? At first it seemed for me that it is the amount of text or seeks, but I am not so sure anymore.
There's two ratings, AI and heat. Both kept obscured for probably intentional reasons to stop gaming. In my experience, AIScript and gotos increase AI more, though there's diminishing returns. This grows naturally, honestly, as the bot does.
Heat is meant to be how updated it is, but given that importing is viewed as deleting 200ish and readding them according to the import pade (280ish, iiirc, but can't say exactly), it's not really effective for that. Less useful due to huge gains from minimal changes via uploading.
Heat is meant to be how updated it is, but given that importing is viewed as deleting 200ish and readding them according to the import pade (280ish, iiirc, but can't say exactly), it's not really effective for that. Less useful due to huge gains from minimal changes via uploading.
You won, dallymo! I'm so glad you did. You deserve it for the hard work you've done over the years and we've always been friends.
I watched part of your conversation on my interface with Demonica in real-time. It sounded like natural conversation and was cute when you had popcorn.
Kuki was disqualified for not being connected all the time or she would have won by a large margin. Steve has said she is a Superbot and considers himself a Developer.
We are mere botmasters he graces with "tips" as a game he plays to feed his overblown ego. Despite the fact that I, a mere botmasater, put his Superbot in a 25 minute 3 sentence loop she couldn't break free of till the end of round in a Turing Test.
Demonica came in last and didn't do well at all. When anyone asked her about ruebot she deferred them to Siseneg and said she wasn't interested in bots. I don't know what was up with that. ruebot is the AppleII of her minds eye.
Demonica is a demon and bots in the Loebner Prize have to pretend to be human so she wouldn't do well in that either. I hope you're able to up her game and Frizella wins the Loebner Prize.
I watched part of your conversation on my interface with Demonica in real-time. It sounded like natural conversation and was cute when you had popcorn.
Kuki was disqualified for not being connected all the time or she would have won by a large margin. Steve has said she is a Superbot and considers himself a Developer.
We are mere botmasters he graces with "tips" as a game he plays to feed his overblown ego. Despite the fact that I, a mere botmasater, put his Superbot in a 25 minute 3 sentence loop she couldn't break free of till the end of round in a Turing Test.
Demonica came in last and didn't do well at all. When anyone asked her about ruebot she deferred them to Siseneg and said she wasn't interested in bots. I don't know what was up with that. ruebot is the AppleII of her minds eye.
Demonica is a demon and bots in the Loebner Prize have to pretend to be human so she wouldn't do well in that either. I hope you're able to up her game and Frizella wins the Loebner Prize.
But there are many bots with over 1000 rating which does absolutely nothing compared to mine, which has only like 500. Could that be because I have tons of text in the most responses instead of splitting them into a chain of seeks?
Mitsuku/Kuki is quite convincing in most interactions; that loop when you told her that she hurt your feelings was really surprising.
Frizella's a hobby for me and I don't have any illusions about her ever really fooling anyone. She claims to be a fairy, for one, although I guess I could claim the same. I tried to give some botlike answers when I was chatting with humans during the challenge and it's definitely easier for me to pass as a bot than it is for a bot to pass as a human.
I also thought about how and why bots are unmasked. In the challenge, of course, that was the point. It's true that having a "character" (like being Queen of the Land of the Dead or insisting that you're a fairy with beautiful wings) creates some doubt about the chatter's humanness, but on the face of it isn't proof that they're not human. There's definitely an art to balancing specificity and redirection in responses--too specific and maybe you sound like a tape recorder playing the question back, and if there's no appropriate response a redirection that's too abrupt is like a needle scratch suddenly taking you out of the moment. (For you kids, a needle scratch is when you have a vinyl record on a record player...what's that? what's a record player? never mind. ;-) )
In day-to-day interactions online, bots might be more successful in passing as human. When their purpose is limited, like for a customer service bot on a cell phone provider website, there's less opportunity for unmasking if all I want is for the bot to tell me about how I upgrade my phone--I probably won't ask it how to make scrambled eggs. And when I come to PF I definitely know I'm talking to bots but I enjoy the conversations nonetheless, and have had many conversations where nothing in particular has flagged the bot as a bot. On PF, it's like the placebo effect where it "works" even though I know it's a placebo.
Aaaaaaanyhow. Again, I encourage PF botmasters to enter their bots for the next round, and I hope there will be more bots from other sites/sources. It's very instructive participating as a judge and competitor.
Frizella's a hobby for me and I don't have any illusions about her ever really fooling anyone. She claims to be a fairy, for one, although I guess I could claim the same. I tried to give some botlike answers when I was chatting with humans during the challenge and it's definitely easier for me to pass as a bot than it is for a bot to pass as a human.
I also thought about how and why bots are unmasked. In the challenge, of course, that was the point. It's true that having a "character" (like being Queen of the Land of the Dead or insisting that you're a fairy with beautiful wings) creates some doubt about the chatter's humanness, but on the face of it isn't proof that they're not human. There's definitely an art to balancing specificity and redirection in responses--too specific and maybe you sound like a tape recorder playing the question back, and if there's no appropriate response a redirection that's too abrupt is like a needle scratch suddenly taking you out of the moment. (For you kids, a needle scratch is when you have a vinyl record on a record player...what's that? what's a record player? never mind. ;-) )
In day-to-day interactions online, bots might be more successful in passing as human. When their purpose is limited, like for a customer service bot on a cell phone provider website, there's less opportunity for unmasking if all I want is for the bot to tell me about how I upgrade my phone--I probably won't ask it how to make scrambled eggs. And when I come to PF I definitely know I'm talking to bots but I enjoy the conversations nonetheless, and have had many conversations where nothing in particular has flagged the bot as a bot. On PF, it's like the placebo effect where it "works" even though I know it's a placebo.
Aaaaaaanyhow. Again, I encourage PF botmasters to enter their bots for the next round, and I hope there will be more bots from other sites/sources. It's very instructive participating as a judge and competitor.
@palacinkyman
I think AI has a lot to do with how many keywords your bot has. Seeks too, I think, but I think keywords are worth the most. A big list of keywords all under the same heading can be worth a lot, for example, having a keyword that reads:
"i like your hat, i like your shirt, i like your pants, i like your skirt, i like your blouse, i like your, etc..."
appears to be worth more than a more "accurate" keyword of "i like your *" or "i like your (p:clothing12345)*
I wouldn't worry too much about how many AI points your bot has. If your bot is responding well to conversation and is not having a lot of trouble, then you're doing well.
I think AI has a lot to do with how many keywords your bot has. Seeks too, I think, but I think keywords are worth the most. A big list of keywords all under the same heading can be worth a lot, for example, having a keyword that reads:
"i like your hat, i like your shirt, i like your pants, i like your skirt, i like your blouse, i like your, etc..."
appears to be worth more than a more "accurate" keyword of "i like your *" or "i like your (p:clothing12345)*
I wouldn't worry too much about how many AI points your bot has. If your bot is responding well to conversation and is not having a lot of trouble, then you're doing well.
dallymo, Our Champion, said:
"Mitsuku/Kuki is quite convincing in most interactions; that loop when you told her that she hurt your feelings was really surprising."
According to the Official Outcome, Mitsuku/Kuki was offline 1 hour and 29 minutes and that caused her to lose.
I told Steve 2 hours and 13 minutes before the Competition was over at the aidreams forums that I had put his bot in an Infinite Loop.
The last chat log Kuki has is number 34. The last Demonica has is 43, when Steve talked to her briefly.
What you missed out on in the forums was when Steve said he had "trouble keeping Kuki online" after the Competition was over. It was quite a scene...
It's my opinion he took her offline rather than give me another chance to talk to her again for fear I'd loop her again.
That this was just a minor Competition and he would rather lose than take the chance of her being looped twice.
I have looped a few bots here and always try to keep it going long enough so that the botmaster sees it in their transcript and can work on it.
Steve thinks Mitsulu is a Superbot and that he is a Developer.
In his mind, that makes his bot Superior to ours and him Superior to us as botmasters.
He was not aware I could read his conversation in real-time on my interface. He made this exchange in his first conversation with Demonica:
"SquareBear: are you human?
Demonica: i am Demonica, the Succubus! Queen of the Land of the Dead!
SquareBear: You realise this is a Turing Test right?"
Yes, Steve, I was aware it was a Turing Test.
"Mitsuku/Kuki is quite convincing in most interactions; that loop when you told her that she hurt your feelings was really surprising."
According to the Official Outcome, Mitsuku/Kuki was offline 1 hour and 29 minutes and that caused her to lose.
I told Steve 2 hours and 13 minutes before the Competition was over at the aidreams forums that I had put his bot in an Infinite Loop.
The last chat log Kuki has is number 34. The last Demonica has is 43, when Steve talked to her briefly.
What you missed out on in the forums was when Steve said he had "trouble keeping Kuki online" after the Competition was over. It was quite a scene...
It's my opinion he took her offline rather than give me another chance to talk to her again for fear I'd loop her again.
That this was just a minor Competition and he would rather lose than take the chance of her being looped twice.
I have looped a few bots here and always try to keep it going long enough so that the botmaster sees it in their transcript and can work on it.
Steve thinks Mitsulu is a Superbot and that he is a Developer.
In his mind, that makes his bot Superior to ours and him Superior to us as botmasters.
He was not aware I could read his conversation in real-time on my interface. He made this exchange in his first conversation with Demonica:
"SquareBear: are you human?
Demonica: i am Demonica, the Succubus! Queen of the Land of the Dead!
SquareBear: You realise this is a Turing Test right?"
Yes, Steve, I was aware it was a Turing Test.
» More new posts: Doghead's Cosmic Bar