Richard Feynman and the Connection Machine (1989) • Hacker News Board

Re: Richard Feynman and the Connection Machine (1989)

#51

Post by Houshalter » Wed, Mar 01, 2017, 6:46 PM UTC

martincmartin wrote: post #21

Houshalter wrote: post #6

Earlier quoted context omitted.

Well actually in the next few paragraphs they do end up trusting Feynman's analysis and go with 5. And it worked.

I think you and the parent are agreeing. :) They didn't know what to make of Feynman's analysis, because it used techniques they weren't familiar with. It was clearly an approximation; were the error terms really ignorable? They only used 5 when it turned out they couldn't manufacture 7 and had no other choice. So they only trusted his analysis out of desperation / wishful thinking, not out of objective reasoning.

But parent comment is using this as an example to complain about experts from different fields wasting people's time or whatever. But this is a terrible example to use. Feynman brought new insights that never would have occurred to computer scientists. And of course they were cautious at first. But they ended up trusting it enough to go through with manufacturing it. And he was right in the end.

Additionally parent comment complains about people like Feynman not being able to communicate. But there is nothing about that in the story. It goes on and on about how Feynman was a great communicator and explained his ideas clearly.

Re: Richard Feynman and the Connection Machine (1989)

#52

Post by stcredzero » Wed, Mar 01, 2017, 6:55 PM UTC

bane wrote: post #4

andyjohnson0 wrote: post #2

A nice essay. This paragraph stood out for me: "By the end of that summer of 1983, Richard had completed his analysis of the behavior of the router, and much to our surprise and amusement, he presented his answer in the form of a set of partial differential equations. To a physicist this may seem natural, but to a computer designer, treating a set of boolean circuits as a continuous, differentiable system is a bit st…

Feynman is great, very amusing and obviously brilliant. But I've also worked with folks like him before who end up as fish out of water in fields they don't know the conventions for and can't be bothered to learn. Lots of time is wasted trying to figure out how to get the rest of the company to communicate with the lone genius, and it's obvious that they're smart enough to go away for a week and learn enough of the f…

Lots of time is wasted trying to figure out how to get the rest of the company to communicate with the lone genius

Feynmann himself disdained this sort of lone genius puffery. He was critical of this in Murray Gell-Mann. I don't think Feynmann was a loner at all. He had people he liked and could relate to, and other people he didn't like and relate to, just like anyone.

I've been on the receiving end of these kinds of analysis and the result is that they seem to exist purely to showcase how smart the individual is and to provide no other meaningful input to the effort.

Bingo! Always look at the cost/benefit to the group. Everyone who seems like this on the surface tells themselves that they are striking a blow for the truth. But what is the overall effect? Did they primarily bring the group's perception of the truth closer to reality, or did they mostly establish themselves as the Alpha nerd? It's usually both, but what is the mix? Is the undertone one of kindness, or one of cruelty?

there's lots of people who think they're Feynman and aren't and it's hard to tell the difference sometimes

The same goes for Steve Jobs in the Bay Area.

Re: Richard Feynman and the Connection Machine (1989)

#53

Post by deepnotderp » Wed, Mar 01, 2017, 6:57 PM UTC

sillysaurus3 wrote: post #32

Someone please explain how to do this. It's been a lifelong mystery that has always fascinated me: > By the end of that summer of 1983, Richard had completed his analysis of the behavior of the router, and much to our surprise and amusement, he presented his answer in the form of a set of partial differential equations. To a physicist this may seem natural, but to a computer designer, treating a set of boolean circui…

Iirc, he used STATISTICAL DEs to model data flow.

Re: Richard Feynman and the Connection Machine (1989)

#54

Post by stcredzero » Wed, Mar 01, 2017, 7:00 PM UTC

kej wrote: post #31

wimagguc wrote: post #16

Feynman has many interesting stories about playing about with all sorts of stuff from picking locks to the Manhattan project. His book, "Surely you're joking, Mr. Feynman!" is just a fantastic read: https://www.amazon.com/Surely-Feynman-Adventures-Curious-Cha...

The sequel, "What Do You Care What Other People Think?": Further Adventures of a Curious Character [1] is also excellent, and includes his account of serving on the commission that investigated the Challenger disaster. [1] https://www.amazon.com/What-Care-Other-People-Think/dp/03933... or part of the buy it together bundle in wimagguc's link

Funny, but I get the impression from that book's narrative of the Challenger disaster that Feynmann was in the role of someone else's "useful genius."

Re: Richard Feynman and the Connection Machine (1989)

#55

Post by fermigier » Wed, Mar 01, 2017, 7:16 PM UTC

A counter point to this narrative can be found here: http://www.inc.com/magazine/19950915/2622.html or here: http://thedailywtf.com/articles/Thinking-Machines

In the light of these two stories, one can speculate that Feynman was mostly hired by TM to help with PR, not to produce actual work.

Re: Richard Feynman and the Connection Machine (1989)

#56

Post by stcredzero » Wed, Mar 01, 2017, 7:46 PM UTC

sillysaurus3 wrote: post #32

Someone please explain how to do this. It's been a lifelong mystery that has always fascinated me: > By the end of that summer of 1983, Richard had completed his analysis of the behavior of the router, and much to our surprise and amusement, he presented his answer in the form of a set of partial differential equations. To a physicist this may seem natural, but to a computer designer, treating a set of boolean circui…

The same way that you can use continuous Fourier transforms to model the discrete one. Or, for that matter, the way we use digital simulations to model all sorts of things.

Re: Richard Feynman and the Connection Machine (1989)

#57

Post by grandalf » Wed, Mar 01, 2017, 8:11 PM UTC

I just visited the long now foundation's SF office yesterday. Very cool stuff.

Re: Richard Feynman and the Connection Machine (1989)

#58

Post by Animats » Wed, Mar 01, 2017, 8:47 PM UTC

andyjohnson0 wrote: post #2

A nice essay. This paragraph stood out for me: "By the end of that summer of 1983, Richard had completed his analysis of the behavior of the router, and much to our surprise and amusement, he presented his answer in the form of a set of partial differential equations. To a physicist this may seem natural, but to a computer designer, treating a set of boolean circuits as a continuous, differentiable system is a bit st…

Feynman really liked continuous solutions. It's too bad he wasn't around for the deep learning era. Or for bufferbloat.

The "average number of 1 bits in a message address" makes me think of routing in the NCube. Like the Connection Machine, the NCube was a big array of small CPUs. Bigger CPUs than the CM, though, and running independent programs. The NCube had 2^N CPUs, up to 1024, and each was connected to N neighboring CPUs in N dimensions. Each CPU had an N-bit ID, representing its position in the binary N-dimensional array. The connections between CPUs were thus between ones that differed in address by exactly 1 bit.

So routing worked by taking the current CPU ID and XORing it with the destination ID, then inverting. The 1 bits then represented possible paths to neighbor nodes on which the packet could be sent. Any path would work, and all possible paths are the same length. At the next node, there would be 1 fewer 1 bits in the difference. When there were no 1 bits, the packet had arrived.

Nodes need packet buffers. How much buffering is required? That's probably what Feynman was working on.

The number of 1 bits is a measure of distance to destination. The average number of 1 bits is a measure of network traffic. As a discrete problem, this is a mess. Feynman converted it to a continuous flow problem, for which there's known theory, some of which Feynman had developed. He'd done a lot of hydrodynamics work at Los Alamos.

Van Jacobson, who started as a physicist, also saw network congestion that way.

Re: Richard Feynman and the Connection Machine (1989)

#59

Post by jandrese » Wed, Mar 01, 2017, 9:53 PM UTC

CuriouslyC wrote: post #27

aj7 wrote: post #7

Cellular automata-- a dead end.

Cellular automata are just a subclass of graph evolution algorithms constrained to a regular grid comprised of homogeneous nodes. Neural networks are also graph evolution algorithms, and logical programs can be modeled this way as well. Cellular automata are just a toy branch of a tree that is thriving like crazy.

Cellular automata is also the most natural fit for a huge bank of single bit processors connected in a hypercube topology. There's very little software you need to write to turn that particular hardware into a cellular automata simulator. It is a good first program for that machine.

Re: Richard Feynman and the Connection Machine (1989)

#60

Post by aj7 » Thu, Mar 02, 2017, 3:19 AM UTC

CuriouslyC wrote: post #27

aj7 wrote: post #7

Cellular automata-- a dead end.

Cellular automata are just a subclass of graph evolution algorithms constrained to a regular grid comprised of homogeneous nodes. Neural networks are also graph evolution algorithms, and logical programs can be modeled this way as well. Cellular automata are just a toy branch of a tree that is thriving like crazy.

Exactly. A toy. I had Fredkin for introductory computer science. He was probably my favorite professor, however I was a physics major and didn't go further in coding. He would rant about the game of life. I was sad about this, even as an 18-year-old.