Hacker News Reader

Comment by pacala | Hacker News Reader

Everything you say is correct. Your acid test is great. And yet, we agree that the machine is stuck.

'Does your startup implement every single feature of Windows / Excel / Postgres / AWS and then some? No? Well, then go back to work and come back when you have something truly innovative.'

There is no charity. 'What you did is fairly incomplete, but the ground you covered so far is promising. Here's a list of challenges. Where would you start addressing them?' Maybe the answer will come 10 years down the road, in some University unaffiliated with the original author. In the current system, there is zero investigation of anything else but well trodden orthodoxy.

View on HN

The computer science equivalent of the "type of progress" I would like to see is ZFS.

When ZFS was announced, it blew my mind. The Sun ZFS team showed that traditional RAID had a veritable laundry list of fatal, nigh unfixable flaws. Various forms of silent data corruption caused by things such as reordered writes, partial writes, lost writes, or bit flips, and the like.

Then they showed how Merkle Trees provide a simple, unified, provably correct mathematical model of data integrity that in one fell swoop permanently eliminates all of the issues with traditional RAID data integrity. Poof. Gone! Reordered writes: detected. Lost writes: detected. Bit flips: detected. ZFS could even protect against malware in the drive firmware. RAID has no hope.

They demonstrated that in their testing, they had uncovered bugs in several commercial RAID adapter cards and even the DMA chip of one particular motherboard. These were real problems that really were eating people's data!

Did they have online ZFS pool expand on day 1? No. Defrag? No. Encryption? Nope. In fact, it took over a decade to add all of the various features and niceties that people wanted.

That doesn't matter though, because on Day 1 they demonstrated a fundamentally better way of storing data.

Stephen Wolfram hasn't really done this. To put things in the above terms, he hasn't managed to even store a file and retrieve it intact yet, and there's no firm indication that this will ever be possible. This makes his system indistinguishable from /dev/null.

Worse, his system may be the type where you can put files in, but when you try to read them back out, the filesystem simply contains an infinite list of files, one for all possible bit strings. I mean sure, technically your file is in there... somewhere. But this is indistinguishable from /dev/random. Also not useful.

Stephen's work is intriguing. He wants a model to simulate space, time, matter and energy. Some desirable properties of such a system, from a layman perspective:

1. The model must be fundamentally simple.

2. The model must be able to produce 'interesting' behaviors somehow.

3. The model must exhibit very puzzling behaviors like 'speed of light is a constant', 'space dilation' or 'quantum superposition'.

He observes that extremely simple rewrite systems, like the infamous rule 30, satisfy 1 and 2, under a provable universality result: all 'interesting' behaviors are Turing equivalent. He then observes that the succession of states of the system could be observed as 'time', that having the fundamental simulation substrate create new nodes on the fly could be observed as 'space dilation', that the speed of generating new nodes is a limiting factor in the same way the speed of light is and that having alternate rules applicable in a given configurations could be observed as 'superposition'. This sounds promising. Let's work out the details. To my layman understanding, it's a fundamentally different approach than what physicists have been doing, extremely successfully, in the past 100+ years.

So far, he's empty handed, and very far from concrete results. He's pioneering the exploration of this space by building the first generation of programming languages, the machine code. Maybe 50 later, someone will build ZFS / explain the '3 but not 4 particle generations', using ML to express their thought process.

He's put some thought in attempting to bridge graph rewrite systems to the Universe's behavior. No measurable results, but no [layman's] obvious way to refute either. https://www.wolframphysics.org/technical-introduction/potent...

He really isn't pioneering much though. He talks about graph automata like he invented them, or that he was the first to see that simple rules could generate very complex behavior.

But he wasn't.

I am not saying he hasn't done anything interesting, but it is actually hard to tell exactly where any original work begins, even for someone somewhat familiar with the area of computing related to graph transforms (me), because in his writing he implies he did so much more than he did. So it isn't clear where his contribution line starts.

It rubs me so wrong I can't read anything he writes anymore.

This may be one aspect of the negative reaction of some physicists too.

That is a fair criticism. The tone is off-putting. Partly because of the tone, I considered NKS a load of pointless drivel for 20 years. Yes, absolutely, he didn't invented graph automata. But he's the one willing to propose graph automata as a model of the Universe, and work out some technical details. Or, at least, he's building tooling to experiment with graph automata in this context such that someone else can work out technical details. Or fail to do so. Or fail to do so and prove this is a dead end.

While I'm not familiar with what kind of graph transforms work is done these days, I would have hoped that if someone re-created some version of Lorenz transforms on top of graph transforms, I would have heard of it. The discussion would be more constructive, or destructive as a manner of speaking, if someone would point out that either Stephen is wrong, or old news, on the handful of technical points he is making.

https://www.wolframphysics.org/technical-introduction/potent...

Reply on news.ycombinator.com