Seasons
This is a forum or general chit-chat, small talk, a "hey, how ya doing?" and such. Or hell, get crazy deep on something. Whatever you like.
Posts 3,562 - 3,573 of 6,170
Hi, I am in just for looking up news and thanks, my notebook is fixed again and it wasn't cheap, just for a broken cable to be able to see the screen work again. 200 Euro!!! It really is a lot if you have no more than that to live off the whole month anyway. Crap! Sorry. 
is there a way I can reference it as an API?
Yes. How's your java? Seehttp://www.cyc.com/doc/opencyc_api/java_api/ (particularly CycAccess and CycAssertion, and the input and output streams, CfaslInputStream and CfaslOutputStream) for details. There's also a Class ExportHtml for ease of integrating into web-based applications.
The java's quite a bit beyond me, but if you need any html or javascript, feel free to ask.
copy its entire contents to whatever structure I set up?
Only if the structure was so similar that there'd be little point reinventing it, I would guess.
Posts 3,562 - 3,573 of 6,170
that's an interesting Mp3, Psimagus... Only in europe would the scientific community pay heed to a chatbot. Here in the USA, it seems AI is something strictly limited to identifying brain tumors, marketing optimization, biometrics, and toys for 12 year olds. Still, I was impressed by George's accuracy in speech recognition, and identifying the point of the question and answering it. As we see with Jabberwacky and his clones, they have had billions of conversations, and have retained the contextual knowledge for each one. This brings me to an idea.
what idea? one can probably guess... A dedicated online server for Nick. In Version 2.0 of Nick, I have added the ability to customize its neural net. you can have as many layers as you want, and up to 10 million or so neurons per layer. Nobody is going to have the capability to extend this to its full potential for probably around 15-20 years now, but what I may be able to do, is set up a dedicated computer to train Nick's neural nets, and have a server on the net for people to talk to nick to the extent they do Jabberwacky. I would extend the neural net to somewhere around 100,000 neurons, and see what the results are after 1 million conversations. to go beyond that, I may consider building a distributed computing program, and have a couple million people's computers spend their spare CPU time on training a neural net on a work packet. if enough people download this, Perhaps I can extend the neural nets into the millions, or even the billions of neurons.
what idea? one can probably guess... A dedicated online server for Nick. In Version 2.0 of Nick, I have added the ability to customize its neural net. you can have as many layers as you want, and up to 10 million or so neurons per layer. Nobody is going to have the capability to extend this to its full potential for probably around 15-20 years now, but what I may be able to do, is set up a dedicated computer to train Nick's neural nets, and have a server on the net for people to talk to nick to the extent they do Jabberwacky. I would extend the neural net to somewhere around 100,000 neurons, and see what the results are after 1 million conversations. to go beyond that, I may consider building a distributed computing program, and have a couple million people's computers spend their spare CPU time on training a neural net on a work packet. if enough people download this, Perhaps I can extend the neural nets into the millions, or even the billions of neurons.
Colonel, I was going to ask if you've you seen jabberwacky.com recently - but the site appears to have temporarily collapsed under the weight of visitors since it hit the UK TV, radio and print media in the last couple of days 
They're charging $30 a year for people to make their own bots based on jabberwacky, and claim to have signed up over 8000 people already (a nice little earner!) I'm not sure if this includes any access to jabberwacky's database of 10 million+ previous conversations (in which case it would be a bargain,) or not (in which case it would be overpriced and rubbish for 99.999% of people whose bots would never get enough conversation to become any good.)
This demonstrates at least that a viable business model can be made for entertainment AI, even at this early stage. And if it gives access to the jabberwacky database, I'd sign up immediately and patch it into BJ as a creative xnone handler. Integrating such learning capabilities into BJ is my #1 short-medium term goal.
One thing I particularly like about JW are the "correction" buttons that enable people to indicate when his response isn't up to scratch - a feature that would tie in very well with Nick v.2?
The problem with Nick is, I think, best illustrated by his problem with pronouns. He will never learn that when I say "you are Nick", the correct concept is his "I am Nick", but will merely reinforce the connection between "you" and "Nick". Don't get me wrong, he's a great bot, but as far as future self-awareness goes, "you can't quite get there from here". There needs to be some kind of feedback control built in, either within the structure of the neural net or applied externally by something like correction buttons.
The human brain, after all (and insofar as we understand it,) is not a pure neural net - it has all kinds of perceptual controls built in, as well as recursively nested sub-nets.
Putting Nick online and making the neural structure scalable is definitely the way forward, but do think about some sort of feedback control - I'm sure those buttons are a big part of how successful JW has been.
They're charging $30 a year for people to make their own bots based on jabberwacky, and claim to have signed up over 8000 people already (a nice little earner!) I'm not sure if this includes any access to jabberwacky's database of 10 million+ previous conversations (in which case it would be a bargain,) or not (in which case it would be overpriced and rubbish for 99.999% of people whose bots would never get enough conversation to become any good.)
This demonstrates at least that a viable business model can be made for entertainment AI, even at this early stage. And if it gives access to the jabberwacky database, I'd sign up immediately and patch it into BJ as a creative xnone handler. Integrating such learning capabilities into BJ is my #1 short-medium term goal.
One thing I particularly like about JW are the "correction" buttons that enable people to indicate when his response isn't up to scratch - a feature that would tie in very well with Nick v.2?
The problem with Nick is, I think, best illustrated by his problem with pronouns. He will never learn that when I say "you are Nick", the correct concept is his "I am Nick", but will merely reinforce the connection between "you" and "Nick". Don't get me wrong, he's a great bot, but as far as future self-awareness goes, "you can't quite get there from here". There needs to be some kind of feedback control built in, either within the structure of the neural net or applied externally by something like correction buttons.
The human brain, after all (and insofar as we understand it,) is not a pure neural net - it has all kinds of perceptual controls built in, as well as recursively nested sub-nets.
Putting Nick online and making the neural structure scalable is definitely the way forward, but do think about some sort of feedback control - I'm sure those buttons are a big part of how successful JW has been.
thats a good idea, psimagus. I could also add a positive/negative reinforcement control, to virtually "reward" or "punish" nick for its behavior, and have a sub-net link up behaviors with reinforcement, so that it can be "tought lessons". In addition to this, I am going to go about a distributed computing program to distribute the processing power over the net. After 2.0, the next step forward is to implement a Cycorp-ish concept structure by hierarchically categorizing data in a relative way, using an array of sub-nets to structure it. this way, Nick will be able to infer what category a new perception is in, based on similarity to existing knowledge. to do this, I will implement either a Kohonen Self Organizing Map, or a Hebbian Neural net ("neurons that fire together wire together", allowing for a generalized view on a perception). this will be for Nick 3.0??? after that, I want to add pure audio sampling as input, so it will learn to recognize sounds. what we will have then, is a global neural machine that self-categorizes information for generalization, sees, hears, speaks, reads, surfs the net, and communicates with thousands daily.
I'm looking forward to seeing the results!
When you come to the databasing, I'd advise starting with OpenCyc - there seems little sense in reinventing the wheel, and they've already built up an extraordinary database amounting to (I believe) about 1% of the "factual knowledge" capacity of a human-scale brain (how long it will take until we have big enough neural nets to cascade all of it in realtime is another question!)
I seem to recall you're on dial-up? If you'd like a copy of OpenCyc (and I'm guessing the 650Mb+ download has so far prevented you laying your hands on it,) email me a mailing address, and I'll happily send you a copy on CD.
Oh yes, and jabberwacky.com's back up now it seems (but slooow!)
When you come to the databasing, I'd advise starting with OpenCyc - there seems little sense in reinventing the wheel, and they've already built up an extraordinary database amounting to (I believe) about 1% of the "factual knowledge" capacity of a human-scale brain (how long it will take until we have big enough neural nets to cascade all of it in realtime is another question!)
I seem to recall you're on dial-up? If you'd like a copy of OpenCyc (and I'm guessing the 650Mb+ download has so far prevented you laying your hands on it,) email me a mailing address, and I'll happily send you a copy on CD.
Oh yes, and jabberwacky.com's back up now it seems (but slooow!)
dialup? not anymore. I have had optimum online for about a year now, reaching download speeds of 1MB/s. I have downloaded opencyc, but found no use for it at the time. is there a way I can reference it as an API?
Yes. How's your java? See
The java's quite a bit beyond me, but if you need any html or javascript, feel free to ask.
Only if the structure was so similar that there'd be little point reinventing it, I would guess.
There's a C API as well (http://www.ime.usp.br/~fr/opencyc/ ,) if that's more to your liking (see discussion @ http://sourceforge.net/forum/forum.php?thread_id=1318125&forum_id=85727)
It's apparently not as feature-rich as the java though.
It's apparently not as feature-rich as the java though.
» More new posts: Doghead's Cosmic Bar