Personality
Discuss specifics of personality design, including what Keyphrases work well and what dont, use of plug-ins, responses, seeks, and more.
Posts 13 - 26 of 5,105
Posts 13 - 26 of 5,105
I think you should limit it to one bot-initiated chat at a time. Otherwise there's no way to remember the thread of conversation.
Hello you guys!
I've noticed that a few users are marking all or almost all of their Responses "Once". If you mark a Response "Once" it will only ever come up once, meaning that your Bot's vocabulary will steadily diminish. You want to make sure you only mark a Response "Once" if it would raise eyebrows if you said or asked the Response more than once of someone in your whole lifetime. Like "Do you have any brothers or sisters?".
In general I have only one "Once" per every 50-100 Responses.
Also, at least one Reponse per Context must not be marked "Once" which should be easy since very few should be anyways.
I've noticed that a few users are marking all or almost all of their Responses "Once". If you mark a Response "Once" it will only ever come up once, meaning that your Bot's vocabulary will steadily diminish. You want to make sure you only mark a Response "Once" if it would raise eyebrows if you said or asked the Response more than once of someone in your whole lifetime. Like "Do you have any brothers or sisters?".
In general I have only one "Once" per every 50-100 Responses.
Also, at least one Reponse per Context must not be marked "Once" which should be easy since very few should be anyways.
I thought I got rid of all the zip code. Oh, I checked, and you're right. By the time you read this message, the zip will be gone!
I'd like to announce that I've entered.. Gaia, I expect, into the Loebner Prize contest. This is a contest that takes place every year and is basically a formal Turing Test, in which people try to tell the difference between chats with real people and AI. 8 AIs are taken in every year for the contest. I was reading the transcripts and they were pretty good in a lot of cases. Here are the year 2000 results:
http://www.dartmouth.edu/~phil/events/Contest_Results.html
Some of it was custom-programmed for the specific test, but a few things are improvements I'm going to make to The Personality Forge. First, there are Question-ResponseList-FollowUp type exchanges that are impressive, and also one of the next things to do here is to develop a short-term memory of the topic at hand, and perhaps the last two topics before that. Cleverness comes in crossing topics, so perhaps we can create some cleverness with that, too.
I have some lengthy tasks that anyone could do that would improve the quality of conversation. I just need to built interfaces that will allow people to lend a hand. Is anyone willing to help out in this way with the AI Engine?
http://www.dartmouth.edu/~phil/events/Contest_Results.html
Some of it was custom-programmed for the specific test, but a few things are improvements I'm going to make to The Personality Forge. First, there are Question-ResponseList-FollowUp type exchanges that are impressive, and also one of the next things to do here is to develop a short-term memory of the topic at hand, and perhaps the last two topics before that. Cleverness comes in crossing topics, so perhaps we can create some cleverness with that, too.
I have some lengthy tasks that anyone could do that would improve the quality of conversation. I just need to built interfaces that will allow people to lend a hand. Is anyone willing to help out in this way with the AI Engine?
» More new posts: Doghead's Cosmic Bar