[8pla.net]

Codename: Young Alpha

This is a young alpha prototype. Not fully  worked out yet.  Before this, I created a more complex version, but it runs in a shell.

On the first try I failed to port the shell version to the web.  So, to rethink my strategy, I started from scratch creating a basic design with the goal of making it compatible with the complex design running in the shell.

If you think you can see where I am going with this, then please comment.  Nothing is getting logged this early in alpha testing.  So, I would appreciate hearing your impressions, directly from you.   Thank you.

Live Demo: http://aihax.com/codename

[LOCKSUIT]

Ohhhhhhh ohhh oh I reallly like this !!! Tell us the reasoning behind this! I myself am working very hard developing a real NLP AGI and would greatly benefit from knowing a short comprehensive explanation about what this is and how it works.

It seems to be directed at yes/no/maybe questions. And the comment below tells you if you were right or wrong. It has some randomness included too I see ex. the backprop answer to us is always 1 type if you say "maybe" > could be / fuzzy / intermediate.

I see that "ok" is a win-win line heh heh. Always correct.

If you talked about your wife named "Bugs", and THEN asked "Do you love bugs?" You will say yes!

If someone says "Do you love to eat cats?" you say "No." (hopefully..), but if they say "Do you hate to eat cats?" you say "Yes." The RNN highlights yes and no by "do you", and uses hate/love and cat/apple to do some weight-ins. You then give your answer yes OR no depending on the weights.

[ranch vermin]

Just keep saying depends to it.

[8pla.net]

I am so inspired by your feedback Locksuit!

The reasoning behind this is actually emerging. For design purposes, it was intentionally lightweight. Not low quality, just minimal for launch.  The main focus initially was to setup a sensible way to simulate a shell over the Internet.  To some extent the power of the shell is being reimagined over the Internet to seek popularity with people.  Young Alpha, you may notice has become slightly more deterministic, thanks to your very useful feedback.

Thank you.

[8pla.net]

Ranch,

Your feedback helped train it on more rules.  Turns out handling "depends" can be applied as synergy.  You may notice the results have since become slightly more successive.

[LOCKSUIT]

Is this the order you got? Correct the list if not.



no
false
wrong
incorrect
true?
why?
never?
neither

yes
true
definitely
correct
either

maybe
fuzzy
could be
intermediate

ok
fair

questions/answers/remarks+randomness

[8pla.net]

The order you posted is just wonderful, Locksuit!

Since your feedback on Wednesday, Young Alpha has grown...

Now, a question mark means Young Alpha is uncertain with your answer.

So, when a response of "true?" roughly equates to... False.

This means Young Alpha is not convinced of the truth.

With, "never?", Young Alpha questioning whether it is really never.

Thank you for your help with this.

[WriterOfMinds]

I had a little chat with Young Alpha this morning. I think the biggest upgrade I would recommend is giving the user an ability to respond to its followup questions. E.g. here is our very first interaction:

Will you help me, please?
>> With what?
Does "With what?", mean yes or no?

If it can only handle yes/no responses, that's fine ... but I couldn't clarify, because it immediately asked me a different question in the top bar.

[8pla.net]

Here are some brand new quotes from Young Alpha with the latest upgrade.  The simulated shell has been upgraded too. 

Will you help me, please?   
>> with what?
Does "with what?" mean yes or no?

Will you help me, please?
>> yes
Yeah!

Can I be your friend?
>> Yes
Affirmative.

Young Alpha, in very early design stages, has yet to be fully implemented.  Thank you for your recommendation, WriterOfMinds.

I think the biggest upgrade I would recommend is giving the user an ability to respond to its followup questions.

I couldn't clarify, because it immediately asked me a different question in the top bar.

[8pla.net]

Oh, just noticed 1 Like Freddy likes this.
Thanks for the Like, Freddy!
Tell us what you Like the most (or the least)?
How do you suggest more people may Like it?

[Freddy]

Yes I popped a like in there as I didn't have time to write a post when i first saw this - but I did have a quick chat with the bot.

I liked that the bot seemed to learn via interaction. I assumed it is building up a knowledge base scored by how many times things were rated and how they were rated.

I had a similar thing going on with Jess - whereby she is learning how much words get used and in what context - but mine is a more passive thing (in terms of what the user does) as she learns as she goes along from what is told to her.

I did plan to have her create sentences from this data and ask a user if it makes sense, but I have not got around to that yet.

So yup, that's why I found things interesting !

Also, are you going to make it so that the bot can state truths it feels confident of in some way ? I mean, how are you going to use the data the bot collects ?

Good work thus far 

[8pla.net]

Freddy,

Young Alpha is programming itself but forgets everything it codes.
Simple to check, really... Just do a view source with each interaction.
There is no log because initially Young Alpha was meant to be temporary.

Initially,  I built a command line version that ran in a shell.
Young Alpha was designed to simulate the interface connections of the
command line version to test porting it to the web.

The command line version runs great stand alone, but has yet to be ported the web.
So, this means for the time being, Young Alpha gets to stay on the Internet.
And the command line version gets to keep running stand alone in a shell.

[8pla.net]

Hey ivan.moony,

Thanks for the Like!

Two Likes... I can't believe it?

Young Alpha was supposed to be replaced it's older sibling.

However, the older sibling is being difficult about running on the Internet.

I look forward to reading some of your feedback on Young Alpha (if you get a chance).

Thank you, Ivan.

[ivan.moony]

Hi 8pla.net

I've tried to teach it "of course" phrase, but I didn't get impression that Young Alpha understood me. Or maybe I didn't understand it/him/her.

Anyway, I like your occasional experiments. Keep up, it is always nice to see something innovative.

[8pla.net]

That's a great idea, Ivan.

"Of course" is another way of saying yes.

I suppose I could hook up WordNet.

However, I want to keep the codebase small for now