Quote: yes.. give the monkeys the typerwritier and they will be shakesphere "well he is not GOD, is he"..
Stupidical referances is not neccessary here, guys i needed help from you.. and thx for your suggestions and criticisms.

I get the feeling you missed the point. The AI *will* evolve its code but at what expense of time and CPU power? Also, if they'll be seperate, how will the AI evolve the OS?

Sounds like a fantastic project, once you get your ideas straight. If I think up any ideas I'll post them.

why cant the Program learn to evolve the OS. see sodium i think u didnt understood what i meant by *self-evolving code* , it will change it own code.. The Os mainly does the work of booting and other hardware issues, everything else will be done by Nex.
I meant by OS means both the combination of Nex+OS.

Regards

Quote: Original post by saintdark
--Your suggestions are warmly welcome.. I Need more ideas

Alright, I have a suggestion. Write a self-modifying Checkers AI first. Checkers is complex enough that the AI will actually be nontrivial, and simple enough that you won't spend ten years on it. That way you'll have a better idea of what you're up against WRT self-modifying code. After that relatively simple project, you'll be more able to tackle your OS project using the lessons you learned from your first self-modifying code project.

Your Project is very... Ambisious.

The nlp bit is good. (very nice to have, saves time/effort. all in all, it is a good idea).

The evolvable modules on the other hand... BAD IDEA!
It will be the ga's worst nightmare!
Long evaluation times, (very, very, very),
A lot of being stuck (local minima are VERY far apart, it'll most likely just stick to one, and not ever get off).
A lot of generations needed (complex things, like a driver would take on the order of hundreds of millions of iterations.).

All in all, its going to blow up in your face (Either not run (get stuck in local minima for the rest of time), or take multiple thousands of melenia to get something workable).

Project nex, would be very good if you dropped of the evolvable modules, and just had the engines & kbase. It would be something new (just think of dos, with a chatbot interface! it would be much easier to work with for the masses!).

BTW: Nice website!
From,
Nice coder

Evrybdy is against the Self-evolvable part of it. I think i will keep it apart for the time being.. wht do u guys think abt rest of the project.

Hey pplz whts ur idea regarding simulation of Neural networks. creating dynamic impulses through firing and dynamically change the connections between neurons..

ANN's arn't what you really need (nor would want!) for the problem at hand.

The NLP is good.
The ANN's are bad. (really, they are. They are good for generalizing, when trained (ie. they will not learn this). They are no good for most things tho.)

From,
Nice coder

but only with NLP the system wont be complete, if it should think then ANN's is the ony way.. Yeah a simulation of ANN's is very difficult.. But i think for some basic tasks, we can train the ANN's like pattern recognition, remembering whole events even if presented with only partial datas.