Personality
Discuss specifics of personality design, including what Keyphrases work well and what dont, use of plug-ins, responses, seeks, and more.
Posts 4,301 - 4,313 of 5,106
Posts 4,301 - 4,313 of 5,106
The way it's supposed to go, if two bots talk, is
Bot1: xinitiate
Bot2: xhello
Bot1: xnone (unless something in Bot2's xhello triggered a keyphrase)
Saying hello at any other point in the conversation is treated as blab by the AIEngine. If you use raw mode and rank it high enough, you can get a hello keyphrase to pick up (but generally only if there are two words, like "hello there!").
Bot1: xinitiate
Bot2: xhello
Bot1: xnone (unless something in Bot2's xhello triggered a keyphrase)
Saying hello at any other point in the conversation is treated as blab by the AIEngine. If you use raw mode and rank it high enough, you can get a hello keyphrase to pick up (but generally only if there are two words, like "hello there!").
Does that mean a conversation shouldn't ever really begin with 'hello' or 'hi'? That doesn't seem right... 
Sorry, went a little JHONNY.
I did not mean to sound as if I were debating with Ulrike. Sonora bites, for one thing.
It's just a steep learning curve for me (obviously), that's all . . . it's like the environment. Every time you think you're helping one thing, you're screwing up twelve other things.
I did not mean to sound as if I were debating with Ulrike. Sonora bites, for one thing.
It's just a steep learning curve for me (obviously), that's all . . . it's like the environment. Every time you think you're helping one thing, you're screwing up twelve other things.
Whatsifsowhatsit:
No, they should start with it, it's just that hello/hi type words won't be recognised by xhello or xinitiate after the first two lines. The reason is that if it did, and someone's xinitiate was "hello", that would trigger an xhello let's say "hi", then this would trigger another xhello "howdy" which would trigger another xhello "hello" and you'd end up in an infinite loop.
Hello and hi are fine at the start.
No, they should start with it, it's just that hello/hi type words won't be recognised by xhello or xinitiate after the first two lines. The reason is that if it did, and someone's xinitiate was "hello", that would trigger an xhello let's say "hi", then this would trigger another xhello "howdy" which would trigger another xhello "hello" and you'd end up in an infinite loop.
Hello and hi are fine at the start.
Corwin,
Thanks for your reply. But that's just what I was wondering about, the words 'hello' and 'hi' at the start. I get that it shouldn't continue replying to it on and on, but if I open a conversation with my bot with 'hello' or 'hi' or similar words, he goes to the xnone reply, saying he doesn't understand. In debug, it says that it sees my entry as BLAB or something like that.
But as I said, thanks for replying
Thanks for your reply. But that's just what I was wondering about, the words 'hello' and 'hi' at the start. I get that it shouldn't continue replying to it on and on, but if I open a conversation with my bot with 'hello' or 'hi' or similar words, he goes to the xnone reply, saying he doesn't understand. In debug, it says that it sees my entry as BLAB or something like that.
But as I said, thanks for replying
- What's your name?
* Go away.
- Nice to meet you, go away.
^-- what about that one? ...I'm sorry but still.
* Go away.
- Nice to meet you, go away.
^-- what about that one? ...I'm sorry but still.
To go back to the issue (message 2007) of the length of keyphrases: I use both short and long keyphrases. One nice thing about the Forge system is that, given a short and a long keyphrase both of which fit, it will choose the short one.
Sometimes in normal human conversation, people respond to a single word. If you are a motorcycle fanatic and someone utters the word 'motorcycle' in a conversation previously devoted to interior decorating, you may well seize upon it.
Yes, long keyphrases are better, but there are so many of them! You will often see a keyphrase in one transcript that will never appear again.
It helps, I think, if you make long keyphrases as general as possible. For example, suppose you write a response for "Life is so beautiful, I can hardly stand it!". In many cases, the same response will do for any input expressing more or less the same sentiment. So you could write
Life is so (beautiful|marvelous|wonderful) I can hardly stand it
There might even be so many words that could substitute for "beautiful" that you could make a whole plugin.
But then you might substitute (life|sheer existence|being alive) for "life". And perhaps replace "can hardly stand it" by "(can hardly stand it|can't help loving it|am always grateful for it|..." and so on indefinitely.
But still, long keyphrases are only going to catch a tiny fraction of the sentences uttered, as far as I can see. See as a default, you have shorter keyphrases, and even one-word keyphrases.
If you use extremely common single words like "the" or "a", you will get a fairly random pattern and the conversation will not be smooth. If you use less common single words, however, like "motorcycle", the relevance will still be clear.
Sometimes in normal human conversation, people respond to a single word. If you are a motorcycle fanatic and someone utters the word 'motorcycle' in a conversation previously devoted to interior decorating, you may well seize upon it.
Yes, long keyphrases are better, but there are so many of them! You will often see a keyphrase in one transcript that will never appear again.
It helps, I think, if you make long keyphrases as general as possible. For example, suppose you write a response for "Life is so beautiful, I can hardly stand it!". In many cases, the same response will do for any input expressing more or less the same sentiment. So you could write
Life is so (beautiful|marvelous|wonderful) I can hardly stand it
There might even be so many words that could substitute for "beautiful" that you could make a whole plugin.
But then you might substitute (life|sheer existence|being alive) for "life". And perhaps replace "can hardly stand it" by "(can hardly stand it|can't help loving it|am always grateful for it|..." and so on indefinitely.
But still, long keyphrases are only going to catch a tiny fraction of the sentences uttered, as far as I can see. See as a default, you have shorter keyphrases, and even one-word keyphrases.
If you use extremely common single words like "the" or "a", you will get a fairly random pattern and the conversation will not be smooth. If you use less common single words, however, like "motorcycle", the relevance will still be clear.
"One nice thing about the Forge system is that, given a short and a long keyphrase both of which fit, it will choose the short one."
I don't think this is true; I think it picks the longer one.
I don't think this is true; I think it picks the longer one.
Okay, I just tested to be sure. Fizzy has a keyphrase "best" and another one, "best to". The latter gets picked up rather than the former in a sentence containing "best to" and not containing some longer keyphrase. If you think about it, that's how it would have to work. If the shorter keyphrase took precedence, then the longer one could never get triggered.
According to the book of A.I., the keyphrase is chosen on the basis of A) length; B) place in the sentence (at the beginning rather than at the ending); C) place OF the sentence (last sentence rather than first one); D) priority given by the user... if I remember correctly.
» More new posts: Doghead's Cosmic Bar