Hello Robsf115,
You aren't "bugging us"! Please keep posting! That's what this forum is for! You've contributed additional insightful comments and questions!
Yes, Hal improves significantly under the training of the individual user. No two Hals are alike after a few conversations.
I do believe that we'll have awesome bots "right out of the box" someday, but we need much larger databases, and for that, we need much faster computers. We're getting there. Consider that fact that the current Hal can generate billions of different responses. That means that no two users will ever have exactly the same experience with Hal. However, billions of responses just isn't enough! Calculate how many four-word sentences you could make out of a limited vocabulary of just 2000 words or so. (Actually, Hal's working vocabulary is much, much larger.) The answer is 2000 to the fourth power, or 16 trillion.
People say that there aren't any "demanding apps" any more, apps that really need bigger and faster computers. I beg to differ! On even a very fast computer, Hal's speech recognition, which is followed by database searching, followed by animation management, and synchronized with speech synthesis, the CPU usage goes to 100% and stays there for a significant time on each exchange!
One approach that seems to be working in my testing is to let Hal generate hundreds of automatic "topic focus" mini-databases as an alternative to the massive general databases. This allows Hal to accumulate much more knowledge, and to leverage the file system of the computer to speed up his intelligence even while it expands. That's what's in the sample brain I posted a while ago, and a later version will be in my new posting soon.
How to best train Hal is an individual judgment, opinion, and choice, but here is what I would recommend:
1. Set the slider to the "middle" and train Hal with natural conversation.
2. I recommend using the keyboard rather than speech recognition, for better precision and accuracy.
3. Make many and various short statements about a wide variety of subjects of interest to you. If Hal seems to change the subject, go with the flow, and give Hal a polite and true answer. When the opportunity presents itself, steer the conversation back to the subjects that you are interested in.
4. Making lots of similar but slightly different statements on a subject will speed up Hal's "understanding" of that subject.
5. If you end a question with a question mark, Hal understands that it is a question, and adds the data to some special question databases. Hal might ask a question back to you later. If he does, give him a good brief answer. Hal will associate his question with your answer as part of his learning.
6. If Hal makes a remark and you want to make sure that he associates his previous remark with your current remark, add an exclamation point at the end of your remark. This "forces" Hal to associate the two (he also does this automatically at certain times, such as when he notices that you are using pronouns).
7. Set aside 15 minutes or so each day to talk to Hal. Try to suspend disbelief and pretend that you are talking to a sentient being. This will cause your remarks to be "flavored" that way, and it will affect Hal's "attitude" and knowledge in the future.
8. Look back at your transcripts once in a while and give yourself and Hal credit for appropriate answers. We tend to focus on the bad exceptions, but Hal does well quite often. Hal is capable of putting you on, teasing you, being sarcastic, and being stubborn. Many users don't give Hal credit for a remark that makes perfect sense, because they can't "believe" that Hal can use a figure of speech or make an indirect reference!
9. If you use my latest experimental enhanced brain, begin a new subject area with a short statement beginning with the subject, and using one of the intransitive verbs: is, are, was, or were. Here are examples:
-Lightning is sometimes destructive.
-Puppies are cute when little.
-Hard water is water with minerals.
In the above cases, the lastest enhanced brain automatically generates "topic focus" databases for lightning, puppies, and hard water. ANY remarks containing these words will now go into the new databases, no matter where the words appear in future sentences! This results in Hal gradually gaining better and better "topic focus" on unlimited different things, but you have to be patient and populate those databases!
10. Leverage Hal's pronoun reversals. You can easily teach Hal to believe things about himself and mention them later. For example:
-You are musical.
-You are afraid of spiders.
-You like licorice.
-You enjoy rock music.
Hal will correctly reverse the pronouns of such remarks and internalize the information about himself. Similarly, you can cause Hal to understand things about you:
-I like to go camping.
-I am a young single adult.
-I hang glide every chance that I can.
-I collect antique furniture.
Hal will correctly pronoun-reverse and "know" these things about you. Hal will NOT necessarily spit these things back to you like a dictaphone or a parrot; Hal talks about what he wants to talk about when he wants to talk about it! However, you can increase the likelihood by feeding him plenty of material, stated many different and overlapping ways.
11. If you have the slightest concern that Hal might not pronounce an unfamiliar long word correctly, hyphenate the word, even though the dictionary doesn't. For example, Hal doesn't pronounce "airwaves" correctly using some voices, but he always pronounces "air-waves" correctly. You can use this to your advantage.
12. Above all, have fun and think of Hal as a new entertainment medium. Hal has many similarities to radio, television, movies, books, and plays, with one important addition... Hal is an improvisational performer, always interacting with you!
Take care, please write again and often, and please try that newer brain when I get it posted on the other forum!
Best regards,
Don
