Re: Syllabus for Seed Developer Qualifications [WAS Re: Some considerations about AGI]

From: Mike Dougherty (msd001@gmail.com)
Date: Thu Feb 02 2006 - 16:58:10 MST


If a 'node' of the AGI brain were a web server in one instance, an SQL
server in another instance, or an embedded API in a cell phone, what
difference would it make to the overall "thinking" process?

If we accept that "real-time" to a human-biological brain, "real-time" to a
human-enhanced brain, and "real-time" to a substrate-independant Jupiter(?)
brain are all very different scales, then it should not matter who fires the
most neurons or processes the most simultaneous threads. What matters (to
me) is that some new idea is contributed to the pool of humanity's
knowledge. Along the way, it will be the developers of AGI that are making
the contribution. 'Success' will be when the AGI makes its first
significant contribution.

On 2/2/06, Charles D Hixson <charleshixsn@earthlink.net> wrote:
>
> On Thursday 02 February 2006 03:44 pm, entropy@farviolet.com wrote:
> > On Fri, 3 Feb 2006, Kevin Osborne wrote:
> > >> Have you looked at Alice, from Stanford University? Still in the
> early
> > >> stages, but it looks like it supports parallelism in a quite
> interesting
> > >> fashion. Check out "Promises". Nice!
> > >
> > > will do; though new/experimental languages have certain advantages
> > > (evolution says they'll be better) they also have certian
> > > disadvantages; like the necessity to write your own API modules for
> > > solved problems (sorts, collections, tcp/ip, unicode, math, remoting,
> > > etc) - this is not to say that any newer-language-of-your-choice
> > > doesn't have some or all of these features; but does in have
> > >
> http://cpan.org/authors/id/D/DC/DCANTRELL/Acme-Scurvy-Whoreson-BilgeRat-1
> > >.0.readme ? :-)
> >
> > One problem with most experimental languages is that they arn't
> compiled,
> > or if are don't have the low level access to the machine that C++ does.
> > I say C++ because the inline methods/functions give you high level
> > abstract code that can compile directly down to an instruction or two of
> > assembly. GCC these days is quite good at peeling away the layers of
> > abstraction. For example an inline getter/setter method usually produces
> > exactly the same code that a direct assignment does.
>
> To me this sounds like "premature optimization". Almost all of the
> languages
> either have or develop an FFI interface that lets one get down to C (not
> usually C++) code, and developing at a higher lever is much faster. You
> can
> often pay a price in speed of execution, but usually a profiler will let
> you
> trace that to small pieces of the code that can then be optimized. (One
> of
> the great disappointments that I had with Pyrex is that even though it
> links
> Python and C in a very nice way, it was specifically announced that the
> developer had no intention in using this to create optimized code. So you
> still need to optimize in C and then link it into Python.) What I'm
> currently experimenting with is Ruby (in my opinion the nicest of the
> languages...but SLOW [so you need a faster machine]). Ruby has a way of
> embedding C code within it (small chunks only, probably, but that's enough
> to
> handle the calls), so when an application gets to the point of optimizing,
> then you can profile & optimize. (AFTER checking your algorithms!)
>
> I've also started wondering about using Apache to handle parallelism. I
> haven't looked into the process at all, so this may be totally
> unreasonable
> (or have unreasonable overhead), but Apache is noted for handling lots of
> requests at once with unpredictable synchronization between the
> requests. A
> lot of things can be served over the web, not just html, and the mechanism
> for dispatching dynamically created web pages could probably be used in
> lots
> of other ways (XML is used to represent entire databases...I generally
> think
> of this as a bad idea, but it would make a dandy method for communicating
> between processes that weren't necessarily physically connected...and
> someone
> else has already *done* most of the work).
>



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:55 MDT