Log in

No account? Create an account
1 Peta FLOPS - The Singularity

> Recent Entries
> Archive
> Friends
> Profile

July 30th, 2006

Previous Entry Share Next Entry
04:32 pm - 1 Peta FLOPS
According to this article, Japan's MDGrape-3 has recently been clocked in at 1 Peta FLOPS, that's 1015 floating point operations per second.

If the human brain really requires 1016 FLOPS to simulate in real time (this seems to be about typical for estimates of this number), then we only have a factor of 10 to go. Three more years perhaps?

(5 comments | Leave a comment)


Date:July 30th, 2006 03:22 pm (UTC)
Isn't the availability of software the real issue? I think we will have the raw computational power long before we have the ability to do anything substantial with it. With the singularity, I think the lag of human understanding and implementation plays a role that is oft ignored.
[User Picture]
Date:July 30th, 2006 03:24 pm (UTC)
That was me.
[User Picture]
Date:July 30th, 2006 04:19 pm (UTC)
I agree that the main problem is knowing what to do with the hardware. That's why my research interest is how to make use of computer power to do intelligent things, not how to make more powerful hardware. Indeed I think more powerful hardware might actually be dangerous. Imagine if nobody figures out the algorithms behind powerful AI for 20 years, during which time the available computer power gets to the level of thousands of times more than the human brain. Then some smart person works out the right algorithm and puts it into a super computer.... all of a sudden the world goes from weak AI to super intelligent AI, perhaps in a matter of hours.
[User Picture]
Date:July 30th, 2006 09:43 pm (UTC)
Rather than some smart person entering the right algorithm, what if our aforementioned smart person writes a program (like the designing GA that gets mentioned every now and again) to design/evolve/work out the right algorithm?

[User Picture]
Date:July 31st, 2006 08:53 am (UTC)

In fact my current preferred approach is to narrow down the structure of the cortex as much as possible and then use a GA to explore this space trying to find the right parameters. Of course if I had 10,000 computers to use it would make the probability of success much higher.

> Go to Top