This one's not an April Fools' joke, I'm happy to say. Over the past two weeks I wrote an extensive technical report on the construction, coding, and performance of Simms AI v1, the first incarnation of my self-trained, conversational neural-network-based bot. It's over a year since Simms AI first went online, and 2 million neural connections later, it was time to evaluate its functionality and start looking towards the next version. It is currently online under the screenname SimmsAI should you wish to chat with it; be patient, it lags sometimes. This report explains how it parses and constructs sentences, how its custom thread-management system is structured, and how Simms performs technically and linguistically. Please check it out below and give me feedback, either on the report, on Simms v1, or on the upcoming Simms AI v2.

View Simms AI v1 Technical Report
Simms AI v1: On the Design of a Self-Trained Conversational Artificial Neural Network

Good report, the implementation details are reallly neat, but I don't think you wanted to say that it's lack of understanding of what it says obviates the need for a new version.

Ah, good call, I was using the opposite of the word that I wanted. Thanks for the catch.

Great job Kerm, I m happy to see that you got this out. The report looks great, I haven't yet read it thoroughly, but I skimmed most of it and thoroughly read the other parts. I think I remember you saying you would publish the Simms software itself along with this report and was interested to see it, and if there is any way I can (I have no experience with this type of programming task) help. Also Kerm, an idea for a good interface that is even more common then AIM, a web based chat of almost any sort would be cool, for instance a web form that would lead to a page that would auto refresh every 5-15 seconds until there was an answer, then display it. I might be able to help write one if you want one as I am certainly interested in chatting with Simms, but have no interest in an AIM account other than that. And finally once again, Great Job Kerm.

Read through the paper; good job. I assume you wrote the paper for a class?

You assume incorrectly; I just wanted to professionally document my project.

Glenn: Or just use AJAX and save all that refreshing. But a good idea nonetheless.

I was thinking something along those lines (an automatic refreshing script), but as I have no AJAX experience, I was thinking maybe a flash file made in flex (Which I got a free legal copy of [and most of you can to]) for the refreshing. Do you want me to make a flash interface for showing this sort of thing?

Flash would suck for just a chat. It would be an annoyance and a waste of bandwidth.

And Kerm already has an AJAX chat, its called SAX. Its over there
<----

Or for a more responsive one, look at the chat source I posted a couple weeks ago here.

Why would flash chat be an annoying waste of bandwidth? The flash chat would basically be the same thing as the ajax chat, just in a different language, it would even look almost the same. I was thinking it would interact with a server-side script that would do back end, it would just be a front that would look nice and could also provide extra features easily (such as being able to submit to any number of different web chats alone or all at once from the same input field, I don't think that you can have multiple submit buttons on a single form field even with AJAX (though since I haven't looked much I am not sure).

On your other comment about SAX, I was thinking of a dedicated one on one chat with the SIMMS server (or an interfacing server), so SAX could probably be modified to do that, but it does not currently do that (at least to my knowledge).

Because SWF files are considerable larger than JavaScript, and they require Flash.



But it wouldn't behave the same unless you put some work into it, and it would require Flash player for no reason whatsoever. And it isn't "just a different language", it is an entirely separate execution environment with its own framework and a bunch of hassle, especially for mobile devices (which your solution eliminates for, again, no reason at all)

I realize you love flash, but it simply is the wrong tool for the job (its the wrong tool most of the time)

http://flashsucks.org/ <- chat definitely falls under the "misuse" column.



Well duh, that is how all chat works. It wouldn't be chat if it didn't talk to a back end (unless it was P2P, but that is unnecessary for chat)



Since you already admitted you don't know AJAX, I'll let this slide. But suffice to say you are dead wrong. And HTML+CSS looks better than Flash 99% of the time.



SAX is simply a method of conducting a chat. Kerm can hook the other end up to whatever he wants. He doesn't need to modify the current SAX implementation, but rather fork it and then build off of that.

Ok ok, let's not get too far offtopic. I vote we leave it at some kind of web interface being a good idea, which seems to be a consensus. Other comments on Simms AI v1 or my thoughts and plans for Simms AI v2?

I'm disappointed by the, errr, distinct lack of haggis in the report.

Hehe, I actually should have mentioned that as a weakness of the algorithm, that it can be forced to highly value a particular word or phrase through repetition.

It's really frustrating to try and train it that way actually, because even after an hour or so of persistent repetition, it doesn't seem to be picking it up, but then coming back to talk to it later its spouting out all over the place.

Unfortunately I cannot read the PDF, it seems to think Acrobat is already running in the background

That's awkward, you should use a less-retarded PDF reader like FoxIt.