Zabaware Forums > Ultra Hal 7.0

Persistent Memory?

(1/2) > >>

mcg71:
I'm new to Hal and admittedly haven't spent a lot of time experimenting with it. However, one thing that has really jumped out at me is that there doesn't seem to be persistent memory. I had the idea of possibly using AI to perform as an independent NPC in a D&D campaign with its own thoughts and contributions, but I keep running into the issue of Hal not remembering what it has been told? Is there a way to signify that information should be retained for future retrieval? For example, if I mention a fictional village and some of its characteristics, I'd like the AI to recall those details as memories.

I've looked briefly at the Brain Editor, and it isn't readily apparent to me if this is something that can be done. Would love to hear suggestions or places where I can learn how to more effectively use the Brain Editor. I've been using Role Plai, a mobile app, and it generates interesting stories, but the memory is not persistent over many lines of text. I was hoping to overcome this limitation with Hal. Is this a realistic expectation?

Art:
Welcome!

Have you experimented with using HAL's Intelligence Slider? While not perfect it can affect learned information.

Have you tried using different Plugins? Some of us have had success using the Markov L5 plugin.

We have also found that Repetition of your statement or question in different forms can help. For instance, Travis is the king of Ramhall. The King of Ramhall is Travis, Ramhall has a King and his name is Travis...and so on.
When doing this, ignore HAL's response and proceed with your next statement.

This seems to give HAL more than one way to respond to future questions regarding the King of Ramhall.

With HAL, some methods work a bit more effectively than others. Have fun with it. Hopefully, some others will weigh in as well.


mcg71:
First, thank you for the response! I did play with the slider, I have it somewhere between 32-35. I also played around with the copy from clipboard option last night. I remember reading somewhere on the forum about people using SQL Lite, so I'll look into that as well. I'll check into the Markov L5 plugin, sounds interesting. So far, the only modifications I've done are cosmetic (Margene face w/lip synch and one of the other voices).

Thanks again!

lightspeed:
also something i have found out seems to be in my opinion the larger the brain gets that adjusting the slider a little higher helps , but you have to do it little by little until you see it staying on subject better . i also was using the markov uhp plugin to as it seems to help. me and art both ran test with it.

Art:
Look through the File Sharing section for a more diverse collection of character heads which are very nicely done and certainly better than the original ones that came with HAL!

If you are using HAL's original brain, you shouldn't need the SQLite Studio program yet. The issue happens when HAL's brain gets large and has many tables other than the original ones that shipped with HAL. The issue then becomes that the HalBrain Editor will NOT show all of the associated tables by using the Brain Editor.
The SQLite Studio editor will show all of the Tables but it does have a bit of a learning curve.
------------

You touched on another way to train Hal...the copy and paste in the input box. You can find any site of interest then drag and drop new info into the input box and press Enter to let HAL respond.

Try the Local Wikipedia for quite a large volume of material for HAL to play with for its responses.
-------------

If interested in the Local Wikipedia plugin, you can find the download link here:
https://www.ultrahal.com/community/index.php?topic=17035.msg86640#msg86640

Navigation

[0] Message Index

[#] Next page

Go to full version