Author Topic: Mr. M.  (Read 17995 times)

Medeksza

  • Administrator
  • Hero Member
  • *****
  • Posts: 1469
    • View Profile
    • http://www.zabaware.com
Re: Mr. M.
« Reply #30 on: July 02, 2012, 06:18:03 pm »
The Ogre project is on hold until I have enough funding to pay the developer to finish the project. We purchased the rights to use Amy but unfortunately different leads for funding the project fell through, but I'm always keeping my eye open for other opportunities.

Work on Hal's brain is moving forward though. I've been spending on average about 10 hours a week working on the next generation of Hal's brain. Not as much time as I would like, but I have a 40-hour a week job separate from Zabaware I have to work at now.

I'm currently implementing a more advanced feedback and tracking system in Hal's brain that tracks which responses in Hal's database gets used the most and which barely get any hits. Over time knowledge with little hits will fade out of Hal's memory and eventually get pruned out. Also based on user feedback to Hal's responses (thumbs up/thumbs down) alias connections are automatically generated or relevance adjusted. Responses that get many thumbs up go up in relevance and to the forefront of Hal's database, thumbs down responses slowly get turned down in relevance and maybe eventually pruned out of the database.

The efficacy of this system won't be apparent until there are thousands of users using it and providing feedback to the central database, but I have high hopes for it. I'll release a beta once I feel it is ready to start testing. Right now I'm experimenting with various algorithms and designing the system to scale to meet demand. If I were to throw a ton of users at it now, I'm sure the servers would fail, it's just not ready yet.

Hal has a lot of very old VB6 code which doesn't scale well to a multi-threaded server environment. I've been rewriting large parts of Hal's brain in C++, PHP, mySQL queries, and Sphinx queries. It's a monumental task. I don't want to make any promise of a release date and fail to meet it. All I can say is it will be ready when its ready.
Robert Medeksza

freddy888

  • Hero Member
  • *****
  • Posts: 1693
    • View Profile
    • AiDreams
Re: Mr. M.
« Reply #31 on: July 03, 2012, 07:41:25 am »
Did the crowd sourced money fall through then or was there just not enough ?

What you are doing with the new brain does sound cool though  8)

Any thoughts on Android yet ?

Best of luck :)

Medeksza

  • Administrator
  • Hero Member
  • *****
  • Posts: 1469
    • View Profile
    • http://www.zabaware.com
Re: Mr. M.
« Reply #32 on: July 03, 2012, 09:43:06 am »
The $2500 raised was used to purchase the Amy model and to hire a programmer and artist. We hit a snag trying to convert the Amy model into a format we could use. More developer time is needed to create a better translator to get that model working. Instead, I had the artist work on the frog model and website at www.charimation.com  Most likely the frog will be ready before Amy. My artist will finish the frog character as a favor to me as I've been finding him other paying work to do. Right now he's finishing up work on animations to help teach English pronunciation for a client. When he's done with that he should have more time to work on the frog.

The centralized server server system I mentioned in my previous post will be the backbone for mobile versions of Hal, including Android. Once the server system is up and stable, I will use www.phonegap.com to make a client app to connect to it. PhoneGap will let me easily write one app to target IOS, Android, and Windows Phone all at once.
Robert Medeksza

NIGE

  • Hero Member
  • *****
  • Posts: 1592
  • Bringing things to life
    • View Profile
Re: Mr. M.
« Reply #33 on: July 04, 2012, 02:46:52 am »
WOW!!!!!!!!!!!!
2 posts in 2 days, that's what we want.
Thank you.
The frog??? Will it be working in Ogre?????

You met Chris Shaw from Haptek.
Did anything eventuate from talks with him????

sybershot

  • Hero Member
  • *****
  • Posts: 787
    • View Profile
Re: Mr. M.
« Reply #34 on: July 05, 2012, 09:26:47 pm »
Thanks for the updates, sorry to hear there are still issues with Amy. I can't wait to see how the new frog character comes out, I'm sure it will come out looking great. Can't beat the work of a grateful artist :)

Quote
Over time knowledge with little hits will fade out of Hal's memory and eventually get pruned out.
I hope this feature does not carry over to the upcoming new desktop version. I can see how it would be helpful for maintaining  server space, but to me it would not be a good feature for the desktop version.

 

Medeksza

  • Administrator
  • Hero Member
  • *****
  • Posts: 1469
    • View Profile
    • http://www.zabaware.com
Re: Mr. M.
« Reply #35 on: July 08, 2012, 07:46:01 pm »
WOW!!!!!!!!!!!!
2 posts in 2 days, that's what we want.
Thank you.
The frog??? Will it be working in Ogre?????

You met Chris Shaw from Haptek.
Did anything eventuate from talks with him????

We've tested converting our frog into Ogre format and we haven't run into the issues seen in Amy, so it should work. The model is complete but we need to finish rigging and animating.

From what I know of Chris Shaw, he is focusing his efforts on an updated iOS version of the Haptek player. Not sure of the timeframe or what his plans are for the Windows platform, but the website of the new player will be http://xprevo.com
Robert Medeksza

Medeksza

  • Administrator
  • Hero Member
  • *****
  • Posts: 1469
    • View Profile
    • http://www.zabaware.com
Re: Mr. M.
« Reply #36 on: July 08, 2012, 08:00:55 pm »
Thanks for the updates, sorry to hear there are still issues with Amy. I can't wait to see how the new frog character comes out, I'm sure it will come out looking great. Can't beat the work of a grateful artist :)

Quote
Over time knowledge with little hits will fade out of Hal's memory and eventually get pruned out.
I hope this feature does not carry over to the upcoming new desktop version. I can see how it would be helpful for maintaining  server space, but to me it would not be a good feature for the desktop version.

This feature is mainly for the "global" brain of Hal. If you visit http://zabaware.com/webhal/ you will see its currently about 6 million sentences. I need a mechanism to let the "cream rise to the top" and let the junk wither away. More so than server space this is meant to increase the quality of the data in the database. Right now twitter conversations are being loaded into that database with little feedback on its quality. I'm hoping once we have thousands of users talking to this version of Hal and providing feedback, it will improve its quality greatly.

Separate from this global/collective learning ability of Hal, Hal will have local learning. Meaning things that Hal learns from you will have the highest relevance when talking with you and will only be used when talking with you. This method will be not changed much from the current learning methods Hal uses. In the windows version you will have the option of actually storing this on your local computer as the current version does now or storing it "in the cloud" if you want to use the same personality on various devices. Mobile versions of Hal will only store learned info in the cloud server.

Let me know if that makes sense or if I can explain something better.
Robert Medeksza

NIGE

  • Hero Member
  • *****
  • Posts: 1592
  • Bringing things to life
    • View Profile
Re: Mr. M.
« Reply #37 on: July 09, 2012, 03:04:35 am »
 8) 8) Thanks for your replies Rob,
Exactly what we want,