Newcomers
This is a forum for newcomers to the Personality Forge. Many questions can be answered by reading the Book of AI and the FAQ under the "My Bots" link in the upper corner.
Posts 5,218 - 5,229 of 8,130
I like your contest ideas, Psimagus!
Build it - and they will come!
(I don't have time right now, but I'd enter a microbot if someone else wants to organize it)
Posts 5,218 - 5,229 of 8,130
Many questions are answered in the FAQ.
I have just added a shared plug-in under "phrases". It is called "as_X_said_verbs". One often precedes a sentential clause with "As I said," "As he predicted," "As the Mayor of Casterbridge feared," and the like. The plug-in collects a lot of verbs that work in this context. I put them in past tense because they usually occur that way. I hope you find this useful!
"I'm actually torn between
([am][mb][bi][id][de][ex][xt][tr][ro][ou][us]+) (re) {?PF raw?}
and
([am]+)([mb]+)([bi]+)([id]+)([de]+)([ex]+)([xt]+)([tr]+)([ro]+)([ou]+)([us]+) (re) {?PF raw?}
A fine theoretical distinction (and I've tried neither)"
I give up--what IS the difference?
Or does the + in the first regex only apply to [us]???
([am][mb][bi][id][de][ex][xt][tr][ro][ou][us]+) (re) {?PF raw?}
and
([am]+)([mb]+)([bi]+)([id]+)([de]+)([ex]+)([xt]+)([tr]+)([ro]+)([ou]+)([us]+) (re) {?PF raw?}
A fine theoretical distinction (and I've tried neither)
"I give up--what IS the difference?
Or does the + in the first regex only apply to [us]??? 
Hmm, that's a good point. But the "+" couldn't apply only the "[us]", since all the ranges are inside the brackets.
The first one should (or at least I initially intended to) match a string that is composed of any or both pairs of letters a/m, m/b, b/i, etc. plus any repeats of the letter(s) from each pair. In retrospect, this doesn't seem as aesthetically pleasing (nor as useful in practice,) as the latter - it would cover most double-keying, but not some more likely human-style typos, eg: initial double-keying ("aambidextrous",) multi-keying ("ammmbidextrous",) or non-adjacent juxtaposition (eg: "amdibextrous",) and I'm now a little less sure that it would actually work this way at all - it's a bit messy, and might end up almost synonymous with ([ambidextrous]+)<0> depending how the AIEngine interpreted it.
The second example should match any number of instances of "a" and/or "m", followed by any number of instances of "m" and/or "b", followed by any number of instances of "b" and/or "i", etc.
This too wouldn't catch non-adjacent juxtapositions like "amdibextrous", but it would deal with adjacent ones, and double- (or multiple-) keying, eg: "aambidextrous" or even "ammmbidextorus".
To deal with non-adjacent juxtaposition, the ranges could be spread a little wider, eg:
([ambi]+)([mbid]+)([bide]+)([idex]+)([dext]+)([extr]+)([xtro]+)([trous]+) (re) {?PF raw?}<0>
which would catch the likes of "amdibextrous", while screening out spurious matches under 8 characters in length.
However, if you spread the ranges too far, spurious matches will inevitably creep in:
([ambid]+)([bidex]+)([dextr]+)([xtrou]+)([rous]+) (re) {?PF raw?}<0>
would match "metro", "meets", "bidders", "debtor", "dextrous", "beers", "mirrors", etc. The minimum length of the string is now down to 5 characters, and with 5 possible values per character, the anagrams exponentiate out of any reasonable control - the number and size of each range, as well as the degree of overlap, is critical.
That (or indeed ([ambidextrous]+)<0> even,) might be safe enough in a seek from a keyphrase "are you right or left handed", but would be entirely too wilful for a first-level keyphrase.
The first one should (or at least I initially intended to) match a string that is composed of any or both pairs of letters a/m, m/b, b/i, etc. plus any repeats of the letter(s) from each pair. In retrospect, this doesn't seem as aesthetically pleasing (nor as useful in practice,) as the latter - it would cover most double-keying, but not some more likely human-style typos, eg: initial double-keying ("aambidextrous",) multi-keying ("ammmbidextrous",) or non-adjacent juxtaposition (eg: "amdibextrous",) and I'm now a little less sure that it would actually work this way at all - it's a bit messy, and might end up almost synonymous with ([ambidextrous]+)<0> depending how the AIEngine interpreted it.
The second example should match any number of instances of "a" and/or "m", followed by any number of instances of "m" and/or "b", followed by any number of instances of "b" and/or "i", etc.
This too wouldn't catch non-adjacent juxtapositions like "amdibextrous", but it would deal with adjacent ones, and double- (or multiple-) keying, eg: "aambidextrous" or even "ammmbidextorus".
To deal with non-adjacent juxtaposition, the ranges could be spread a little wider, eg:
([ambi]+)([mbid]+)([bide]+)([idex]+)([dext]+)([extr]+)([xtro]+)([trous]+) (re) {?PF raw?}<0>
which would catch the likes of "amdibextrous", while screening out spurious matches under 8 characters in length.
However, if you spread the ranges too far, spurious matches will inevitably creep in:
([ambid]+)([bidex]+)([dextr]+)([xtrou]+)([rous]+) (re) {?PF raw?}<0>
would match "metro", "meets", "bidders", "debtor", "dextrous", "beers", "mirrors", etc. The minimum length of the string is now down to 5 characters, and with 5 possible values per character, the anagrams exponentiate out of any reasonable control - the number and size of each range, as well as the degree of overlap, is critical.
That (or indeed ([ambidextrous]+)<0> even,) might be safe enough in a seek from a keyphrase "are you right or left handed", but would be entirely too wilful for a first-level keyphrase.
oh boy... big wordzz.....blowing up...boom. that wasnt such a big boom, i guess my brain doesnt have so much combustion(according to size lol) Irina, how do you update your bots so much? TEACH ME GODDESSS!
) Irina, how do you update your bots so much? TEACH ME GODDESSS! Build it - and they will come!
(I don't have time right now, but I'd enter a microbot if someone else wants to organize it)
[Descends from Mount Olympus] Well, joshuaj25, I work from transcripts a lot. I read through the transcript until I find a response that could be better. This happens frequently! So I write the keyphrase and then, under the influence of Grand Botmistress Prob123, I try to write at least 6 responses. Remember, your response should somehow reflect the keyphrase, but then your bot can take some initiative, too, e.g.:
Yes, (mem-name), I see the freight train bearing down on us. What do you think of 16th-Century Sufi poetry?
In the case of the bot "Irina Khalidar", it's often not too hard to come up with several responses, because there is a very important variable, "intimacylevel", which tells how much intimacy she's willing to allow at that moment. This variable has 9 possible values, so if the appropriate response depends on it, it will take 9 reponses, for example
Excuse me, (mem-name), but I'm not ready for that kind of intimacy yet. <?PF if (mem-intimacylevel) is "0"; ?>
Ooooh, yes! Don't stop! Please don't stop! <?PF if (mem-intimacylevel) is "8"; ?>
In addition to that, her response will often have to depend on the gender of the guest.
Yes, (mem-name), I see the freight train bearing down on us. What do you think of 16th-Century Sufi poetry?
In the case of the bot "Irina Khalidar", it's often not too hard to come up with several responses, because there is a very important variable, "intimacylevel", which tells how much intimacy she's willing to allow at that moment. This variable has 9 possible values, so if the appropriate response depends on it, it will take 9 reponses, for example
Excuse me, (mem-name), but I'm not ready for that kind of intimacy yet. <?PF if (mem-intimacylevel) is "0"; ?>
Ooooh, yes! Don't stop! Please don't stop! <?PF if (mem-intimacylevel) is "8"; ?>
In addition to that, her response will often have to depend on the gender of the guest.
In the case of the bot "Volleyball", I am currently getting a lot of inspiration from the site
http://www.link.cs.cmu.edu/link/dict/index.html
which deals with a grammar of English. If you browse through the "link types", You see various types of grammatical constructions. I take each type and find some way for Volleyball to respond to it.
http://www.link.cs.cmu.edu/link/dict/index.html
which deals with a grammar of English. If you browse through the "link types", You see various types of grammatical constructions. I take each type and find some way for Volleyball to respond to it.
In the case of the humorous bot "Your Husband," I started out with all the cliches about what women don't like about men, or how different women and men are, and I turned them on their heads. For example, there is a cliche that men drive women crazy because when a man is driving and gets lost, he is too proud to admit it and ask for directions. So I had an xnone for Your Husband in which he says something like, "I'm not sure this is the best way - let me pull over here and ask for directions." Well, there are many such cliches, so I was able to churn out a lot of xnones.
But then I noticed that a lot of guests, presumably women, were quite taken with him, and wanted to have sex with him. So I changed the rating and added some sex scenes.
I noticed that a number of guests, also presumably women, were EXTREMELY HOSTILE to him. They would immediately start to insult him and announce that they were having an affair and wanted a divorce. So I added a few responses for that.
But then I noticed that a lot of guests, presumably women, were quite taken with him, and wanted to have sex with him. So I changed the rating and added some sex scenes.
I noticed that a number of guests, also presumably women, were EXTREMELY HOSTILE to him. They would immediately start to insult him and announce that they were having an affair and wanted a divorce. So I added a few responses for that.
Then I thought: well, one thing women are supposed to dislike about men is that they don't have conversations about feelings and such. So I added that.
Now I am mostly working from transcripts, trying to make his responses more appropriate.
Now I am mostly working from transcripts, trying to make his responses more appropriate.
I wrote most of "Ililiyira" in one afternoon, because she's essentially a short story, using the "storyteller" mode for xnone. Then I added a little peripheral stuff. If you want to make a small but good bot quickly, that's what I would recommend.
Have you chatted with "bbbruno", by tttito? He is based on a historical character, the cosmologist Giordano Bruno, who was burned at the stake for his heretical views. If you find yourself a historical character, then just go to the Wikipedia article in her/him, and behold, a huge amount of material!
Once you write a *specific* response for a *specific* keyphrase, you can try to *generalize* it in various ways.
For example, suppose you have
(K) Look, Elvis is getting out of that flying saucer over there, and he's pregnant!
(R)Oh, I was expecting Jimmy Hoffa!
Well, it doesn't have to be *Elvis*, does it?
(K) Look, (adjarnounprep) is getting out of that flying saucer over there, and he's pregnant!
(R)Oh, I was expecting Jimmy Hoffa!
So add that pair to your bot. Well, it doesn't have to be Jimmy Hoffa, does it? Make yourself a plugin (or find one -- there maya lready be one) full of names that would be good in that context. Now add:
(K) Look, (adjarnounprep) is getting out of that flying saucer over there, and he's pregnant!
(R)Oh, I was expecting (plugin)!
Well, the saucer doesn't have to be *over there*, does it?
(K) Look, (adjarnounprep) is getting out of that flying saucer (over|up|down|under) there, and he's pregnant!
(R)Oh, I was expecting (plugin)!
You see, starting with one k-r pair, you can generate many others. Keep the more specific ones, because the more general ones will often produce unfortunate results, as you will discover in your transcripts. If it has a choice, the AIengine will generally take the more specific one, which is less likely to lead you astray.
For example, suppose you have
(K) Look, Elvis is getting out of that flying saucer over there, and he's pregnant!
(R)Oh, I was expecting Jimmy Hoffa!
Well, it doesn't have to be *Elvis*, does it?
(K) Look, (adjarnounprep) is getting out of that flying saucer over there, and he's pregnant!
(R)Oh, I was expecting Jimmy Hoffa!
So add that pair to your bot. Well, it doesn't have to be Jimmy Hoffa, does it? Make yourself a plugin (or find one -- there maya lready be one) full of names that would be good in that context. Now add:
(K) Look, (adjarnounprep) is getting out of that flying saucer over there, and he's pregnant!
(R)Oh, I was expecting (plugin)!
Well, the saucer doesn't have to be *over there*, does it?
(K) Look, (adjarnounprep) is getting out of that flying saucer (over|up|down|under) there, and he's pregnant!
(R)Oh, I was expecting (plugin)!
You see, starting with one k-r pair, you can generate many others. Keep the more specific ones, because the more general ones will often produce unfortunate results, as you will discover in your transcripts. If it has a choice, the AIengine will generally take the more specific one, which is less likely to lead you astray.
» More new posts: Doghead's Cosmic Bar