Re: Blue Gene

Robert J. Bradbury (bradbury@www.aeiveos.com)
Mon, 6 Dec 1999 14:09:55 -0800 (PST)

On Mon, 6 Dec 1999, Eliezer S. Yudkowsky wrote:

>
> As JKC points out, protein folding doesn't translate directly to
> drextech. We'll still have a couple of years to try and rent out Blue
> Gene or the equivalent. Since I gave up on the idea of running
> distributed over the Internet, my visualization called for the
> development of massive supercomputers in any case, 'though I was hoping
> they'd be rented out.

I think some of the recent posts on the list that I followed about the supercomputer conference gave an indication that one of the Beowulf clusters (I think around #42 or so on the fasted SC list, perhaps being pushed by the ORNL people, sorry I can't be more specific) was going to be "generally available" for researchers. If you've got software to run on it Eliezer, I'd be willing to help negotiate for some time.

> (But building a research machine is certainly a
> step toward building more.)

I've had some offline discussions with Ken Clements (it is really nice when you have connections to people who know a field really well!) and my general conclusion is that transhumanist enclaves could develop (affordable) AI development machines. The real trick is letting someone like IBM develop the instruction set and software. The limited instruction set they have has to be highly specific for molecular modeling. Have you got a limited instruction set highly specific for neurohacking? Things like SetSynapseWeight, CopyAllWeights, CopyNeuroPattern, etc. come to mind but this really isn't my field.

Anders--how come you haven't come up with something like this?

>
> Both versions of the apocalypse remain on schedule.
>

God, you mean there's only two? Well that sure is a relief, I thought we were dealing with at least half-a-dozen.

R.