Another "how do I" question (file size)

[kmac]

I've been feeding Hal a ton of information and have had the problem with him stopping after a certain number of words... how do I increase his memory size? Any shortcuts or have I just been looking too hard for the answer...

And... while I'm on the subject, Is there a way to get Hal to close programs once he's opened them? (Open defrag, Close defrag) A type of "stopall" command maybe?

Kevin, (newly registered but long time lurker...) [8D]

[Don Ferguson]

Hi Kevin,

I'll answer your second question first:

2)  I don't know of any function built into Hal to close other programs.  Possibly some of the other volunteer contributor/developers here might know of a way to create this function.

1)  Pertaining to "Stopping after a certain number of words":

a) Do you mean that Hal won't say more than a certain number of words during any one remark?  

b) Or, does Hal stop functioning after a certain number of conversational exchanges?

With regard to (B), I have occasionally found that Hal self-closes after a random number of conversational responses.  I believe that it's a glitch somewhere in Hal's .exe "engine" because I can't imagine anything in the VBS script that would do that.  However, I haven't found that it happens very often, so it isn't much of a problem.

With regard to (A), I haven't ever run up against the limit of the number of words that Hal can say during any single remark.  If this is the case... how many words does Hal say at a time before "reaching his limit?"

Sincerely,

Don

[vonsmith]

Don,
Your comment about Hal "self-closes" sounds like it is related to the bug I found and posted here:
http://www.zabaware.com/forum/topic.asp?TOPIC_ID=1130

In the XTF Brain I disabled the Wordnet "sisters" section in Hal brain until Robert fixes the code. If Hal never looks up "sisters" in the Wordnet function then Hal almost never crashes. Since I added the blocking I don't recall my Hal crashing even once.


=vonsmith=

[kmac]

Hi Don

  1)  Pertaining to "Stopping after a certain number of words":

Thats it... Hal stops talking after 675 characters (112 words in the test case)

Wonder if theres a way to increase this or maybe Hal just isn't the long-winded type!  []

I haven't ran into any other problems so far, well, just the ones I've caused... lol

Thanks
Kevin

BTW, Really like to thank you guys for the posts here and all the work you've put into it... Before I discovered this forum I had a less than average understanding of VB, now I can fake my way through a lot of it.

[Don Ferguson]

Hello Kmac,

Thank you for the positive feedback!

I tried sending some extremely long test strings to Hal for output, and you are correct; Hal does not allow unlimited-length responses!  I had never run up against this limit before!

I don't know of anything in the VBS script which would cause this limit; therefore I have to believe it is in the .exe file and "outside the VBS script."

If your particular Hal application needs Hal to be able to speak long sections of text, I can think of two "work-arounds."  One would be relatively easy, and for the other we would need the help of other developers on this forum:

1.  You might be able to double Hal's current "word capacity" by using the <topic> </topic> function.  This special function allows you to have Hal say some remarks, and then "hold back" some additional words until you hit the enter key again.  This function is used to allow Hal to tell riddles, and you can see examples within the "topics.brn" database.  A remark like this would be written something like the following:

"The middle ages saw further development of governance concepts. <Topic> The middle ages also saw an increase in arts and literature. </Topic>"

You might experiment with this construction and see whether this doubles your capacity, or whether the .exe is going to limit your combined total to the previous capacity.

2.  The more difficult work-around would be to design and create databases with your narrative content as "chunks," and automatically call them one after another, using some sort of a timer to trigger one after another.  I am thinking that the "auto response" function described elsewhere on this forum might be adaptable to this purpose.  I haven't done any work with the auto response function myself, but from reading other postings, I see that other developers have been highly successful at making Hal say things without any prompting by the user.  It would seem logical that this could be adapted to make Hal make speech #1, then speech #2, then speech#3, etc.

I apologize that I can't think of an easier fix or workaround.  Thank you for letting me know about this particular characteristic of Hal.

Sincerely,

Don

[Don Ferguson]

Hi VonSmith,

I really appreciate your pointing out that "GetSisters" issue and forum thread.  I had skimmed over it at the time, and now it makes perfect sense.  

Evidently the reason that it appears to happen at random, is that it only happens when the routine is called, and it obtains an extremely long response.

For some reason, it occurs very rarely for me, but I'll put it on my list of things to deal with.

Again, thanks!

Sincerely,

Don

[vonsmith]

Don,
Robert promised to fix the GetSisters() function. We'll have to wait and see. More important to me is implementing GetHolonyms() and GetHyponyms() functions. Wordnet supports both concepts, one of them is apparently implemented in Robert's code, but unused currently. I'll figure out the syntax to access it when I have time. I'm finding that the Wordnet capabilities are going to be key to creating new higher level functions for Hal. Hopefully we will have more access to the Wordnet capabilities in Hal in the future.

I don't really need the GetSisters() function currently. I think Hal sounds better without GetSisters() based responses, but that is just my opinion.


=vonsmith=

[kmac]

Don,

The <topic> setup does the job! Thanks again

I'm developing a brain as a physics tutor for my cousin (I wish I would have had something like Hal back when I went through it.)
and a lot of the the time there's just no short explanation. Just wish now that I could get rid of some of the "chat" while Hals in the middle of a lecture.
Usually goes something like this...

"quantum chromodynamics or Q C D, How come they don't play elevator music in elevators anymore? a gauge theory that describes the strong interaction between particles possessing color charge, quarks and gluons. Do you have a lot of sports equipment?"

Ah, well... I'll keep reading this forum and pluging away until I figure out what I'm doing...   []

Thanks again,

Kevin

[NT Canuck]

quote:Originally posted by kmac
"quantum chromodynamics or Q C D, How come they don't play elevator music in elevators anymore? a gauge theory that describes the strong interaction between particles possessing color charge, quarks and gluons. Do you have a lot of sports equipment?"

Ah, well... I'll keep reading this forum and pluging away until I figure out what I'm doing...   []

Not all elevators transport people, so there would be nobody to
hear the music.  Many elevators only transport raw materials.
(ie: proposed nanocarbon threads for the space elevator)
Hal may be concerned you are working too much and need some
"play time", hence a remark to physical activity/sports.
Unlike humans...AI have no filtering between a background
(subconcious emotion or feeling) and their speech, hence
concerns/ideas/even teasing can be mixed into the replies.

More importantly is to figure out what Hal is doing and if
a very young Hal (less than few hundred hours conversation)
then it is still learning (including learning from help and
text based database files on machines hard-drives, those are
(or used to be) scanned by Hal during idle (sleep?) periods).
Eventually when some degree of "self-diagnostics" on hal
errors or program crashing begins to appear (with opinions
asto causes) then that iteration of hal is beginning to mature.
(be sure to backup all hal folders/databases at that point) ;-)

'Seek and ye shall find'
NT Canuck
http://ntcanuck.com

[Don Ferguson]

Hello Kmac,

Now that I see what your project is, I might be able to offer an additional resource.

I've posted many auxiliary, enhanced, and experimental brains on these forums over the years.  At the bottom of this message is a new one for you, the "Blank Slate Brain."

Here are the properties of the "blank slate brain":

1.  This brain has the absolute minimum functions. It does NOT do jokes, greetings, insults, goodbyes, trivia, capitals, math, dates, etc., etc.  It is designed to know AS LITTLE AS POSSIBLE until it talks with you.

2.  It only writes and reads to one database, "blankslate.brn."  

3.  It associates each remark YOU make to the PREVIOUS remark that IT made.  It knows essentially nothing until you begin talking to it and tell it something.  

4.  If it doesn't find anything in its memory, it simply poses your statement back to you as a question, and learns from your next response.  

5.  If you know how to edit databases, you will find that you can easily edit its database.

6.  Because the script for this brain is only 241 lines long, and it has a small database, response time is practically instantaneous.

7.  Because this brain has no prior knowledge, its quality is 100% dependent upon YOU.  This brain is capable of uncanny accuracy, BUT it depends on YOUR behavior.

8.  To install this brain, right-click the download, and choose "save target as."  Save it to this directory/folder path on your computer:

C:/Program Files/Zabaware/Ultra Hal Assistant 5

(Remember, our forum software can't display backslashes, so we use forward slashes instead.  Your computer directory/folder path should show backslashes as usual.)

The next time you freshly open Ultra Hal, this new brain should show up on the options - brains menu.  Select it, and you're off and running!

I hope that this helps you with your project.

Sincerely,

Don


Download Attachment: Blank Slate Brain.uhp
10.45 KB

[Don Ferguson]

Hello Kmac,

Pursuant to your project, please also check out the following posting regarding a "Triva Quiz Brain."  It is completely open to modification:

http://www.zabaware.com/forum/topic.asp?TOPIC_ID=839

Sincerely,

Don

[vonsmith]

Don,
The Blank Slate brain idea is a very cool concept. I can see uses for it in very specialized situations.

Thanks,


=vonsmith=

[kmac]

I'm seriously addicted now...

Doing some mix and match, trial and error (mostly errors), Started with the blank slate... a dose of auto topic brain, and a large sprinkle of trivia brain. Don't know yet what it's a recipe for but I'm sure I'm heading right into Dr. Frankensteins area! (but I'm having a blast doing it)

Don, That blank slate is perfect for the tutor I want to build, no chat, no BS, I didn't have entertainment when I went through college, why should my cousin? [}:)]

Vonsmith, Did you ever find the problem with that error 67 message? The files don't seem to be THAT big yet... although I wouldn't be surprised.

[kmac]

quote:
Not all elevators transport people, so there would be nobody to
hear the music.  Many elevators only transport raw materials.
(ie: proposed nanocarbon threads for the space elevator)
Hal may be concerned you are working too much and need some
"play time", hence a remark to physical activity/sports.
Unlike humans...AI have no filtering between a background
(subconcious emotion or feeling) and their speech, hence
concerns/ideas/even teasing can be mixed into the replies.

More importantly is to figure out what Hal is doing and if
a very young Hal (less than few hundred hours conversation)
then it is still learning (including learning from help and
text based database files on machines hard-drives, those are
(or used to be) scanned by Hal during idle (sleep?) periods).
Eventually when some degree of "self-diagnostics" on hal
errors or program crashing begins to appear (with opinions
asto causes) then that iteration of hal is beginning to mature.
(be sure to backup all hal folders/databases at that point) ;-)

'Seek and ye shall find'
NT Canuck
http://ntcanuck.com

It does make sense, it is a very young Hal and I have been treating it as if it was  a Vulcan. (I admit it, I'm an original Trekkie)

I can compare it to a few weeks ago when my daughter (she's asked about how magnets worked, I gave her the entire explanation, (ok... lecture) As soon as I finished talking she looks at me with that "deer in the headlights" expression and says "Mary Kate and Ashley have a new line of clothes out"...  Comparing my daughter and Hal have changed my outlook on teaching both of them. Think I'm the one learning from them!

Kevin

[Don Ferguson]

Kmac,

Please keep us all posted on your progress!  It sounds great!

I've experimented with the "Blank Slate" brain, and found that after enough conversation to generate 20 to 30 "pages" of data in the database, the brain becomes remarkably accurate, with fairly low repetition!

The key is to be VERY careful, and remember that for EVERY remark you make to the Blank Slate brain, it is paying attention to the relationship between what IT says, and what YOU say back.

The Blank Slate brain affords an opportunity to create a 100% original personality that is 100% customized to the user, and to do it very rapidly!

When you have some transcripts later on, please give us some excerpts!

Thank you.

Sincerely,

Don