Slashdot Mirror
Beowulf Pioneer Lured From Cal Tech to LSU
An anonymous reader writes "Thomas Sterling, a pioneer of clustered computing, including /.'s beloved Beowulf cluster, has has accepted a fully-tenured professorship at Louisiana State University's Center for Computation and Technology, ditching his old post at Cal Tech. From TFA: "At LSU, he hopes to develop the next generation of high-performance computers that will give birth to true artificial intelligence. By making computer chips more efficient, Sterling believes he can change computing by "one to three orders of magnitude" that will transform how humans interact with technology.""
I think for now he'd better focus on developing sea-water powered computers :)
Image a beo.... oh fuck it. Nevermind.
...was last seen moving northeast towards Mississippi at a brisk pace. Sterling should adjust his trip accordingly.
A nerd improves his chances of getting laid...
EvilCON - Made Famous by
FTA: [Sterling]:"We'll finally stop interfacing with a computer with a keypad," he said. "It's a truly science fiction dream of talking to computers and computers talking back to you."
Great, like I need my computer talking back to me -- I'll be getting enough sass from my teenage daughters by then.
"Trolls they were, but filled with the evil will of their master: a fell race..." -- J.R.R. Tolkien on Olog-hai
At LSU, he hopes to develop the next generation of high-performance computers that will give birth to true artificial intelligence.
2theadvocate was down when I tried to read their story, so mirrors please?
I'll comment briefly (WRTFA):
I am sick of the term next generation: it irks me. I think if you're talking about devoting the next twenty years towards developing true AI, then the focus has to be about the direction that could be taken, the nuts and bolts of it all, and what the setbacks could be. High performance computers are like high performance people, in many ways, or at least they should be. Incentives must exist for a metrological system to present itself into the true nature of self and this measure supercedes the facility of overexaggeration, to the point where no truly defined system can surpass the narrow view of purpose devoted by the creator, without being heralded as a foolish endeavour. The heavy processing of high performance computing works against the nature of AI.
True AI means that mistakes will be made by the creator and the subject, and emotions will exist in the subject to counter-attack development stumbling blocks, and assist in development, or improve development of wisdom and ultimate self-awareness comes only from experiences of contrast, pain and pleasure (for example). These precepts have never come into cause with a system yet, because each system is built as an object and not a person; each system is built for a financial purpose and not a scientific purpose.
Science and finance are enemies, strange bedfellows that hate eachother but rely on eachother, in a bad marriage, with nothing to lose and at times everything to lose. How can balance come to this nature, to enable true AI to come forward out of the ashes?
How is it possible at all? I don't see it. I see just another generation of the same thing, so perhaps the term next generation is apt?
The dangers of knowledge trigger emotional distress in human beings.
Real estate is probably going to be cheap.
You can throw as much hardware as you want at the "problem" of AI, but in my opinion, that isn't the easiest route to achieving a breakthrough in AI - it would be like throwing hardware at a dog's brain - the dog would still think like a dog, only 1000 times faster. Sure, you might see improvement in "mechanical reasoning", and chess playing programs and the like, where most of the neccessary conclusions can be reached mechanically (mathematically), but that's about as far as it will go, I think. You won't get the dog to reach non-doggy (for example, human) conclusions by doing that.
The real key to AI lies in software, and superior algorithms. So far in AI, most of the progress has been on the mechanical side - expert systems using algorithms to match and discard possibilities until it finds the "correct" option. This is a good way of doing things for applications that expert systems are currently being utilized for, but to progress to the realm of true (self-aware) AI, scientists need to find out how it works in biological structures first. Once that has been established, computer scientists can try converting those (theoretical) signals into instructions, and plug those into new-generation algorithms.
Liberal Ontarians and French Quebecers are draining Western Canada's wealth. Stop them now! Support Western separatism.
i know its hopeless..but,
his work these days centers around efficiencies of access gained by putting the dram and processing elements on the same die. partially removing the serialization associated with the standard synchronous memory interface. The architecture also plans on using MTA-style threads to hide latency and increase concurrency.
citeseer
Imagine what we could do with a cluster of these guys!
----- In Your Cubicle No One Can Hear You Scream...
The illusion of being backwater is now trumped by the reality of being underwater.
Apu: I came here shortly after my graduation from CalTech: Calcutta Technical Institute, as the top student in my graduating class of 7 million.
There is another kind of evil which we must fear most, and that is the indifference of good men. -- Boondock Saints
Skynet will be corrupted by Mardi Gras and thus decide to save all the hot chicks but kill the rest of us.
Do really dense people warp space more than others?
...will be one of those rare above-water ones.
Hell, if I wanted to change the performance of my computer by one to three orders of magnitude, I would just run Vista.
Oh, wait, maybe he meant one to three orders of magnitude faster. My bad.
A republic cannot succeed till it contains a certain body of men imbued with the principles of justice and honour.
Allow me to clear up your thinking. Consider Proteus. It is a high-performance simulator written at MIT for MIPS. Some graduate student at LSU ported it to SPARC.
This work is stunningly brilliant and egalitarian.
In the late '80s and early 90s, the eggheads at MIT and Stanford felt that they need only develop simulators for their clique-ish processor: MIPS. Yet, the rest of the world was using SPARC. In this way, the eggheads cornered multiprocessor research for themselves.
LSU actually opened up multiprocessor research to the rest of the world by building a simulator that actually runs on the SPARC machines.
To be fair, I should note that a small team at Stanford did the same thing with ABSS, another simulator that runs on SPARC machines.
His is a predictable move. If after 9 years at Caltech he was still mired in an untenured, non-tenure-track position of "faculty associate", it's natural that he jumped at the chance of becoming a full professor at LSU.
This comment is neither an endorsement nor an attempt to disparage the guy's technical merits, as I don't know the politics going on at Caltech. At least in computer science at Stanford, getting tenure has gotten ridiculously unlikely in the last several years.