This topic is, unfortunately, inextricably inter-twined with AI, so I could as easily be talking about one axis of the problem, as the other. But let’s start with the data centers.
Stderr
All the unbuffered error messages
This topic is, unfortunately, inextricably inter-twined with AI, so I could as easily be talking about one axis of the problem, as the other. But let’s start with the data centers.
The great machine of scientific enquiry continues to grind relentlessly along, in spite of the smoking wreckage made, then thrown in its path, by the likes of RFK, Jr. Eventually, I suppose, the republicans will realize that ice-picking the tendons of American Science just means that some other detestable part of the world will “get ahead of us” in their endless imaginary fight for dominance.
There are a few things I find fascinating about the current war. First and foremost, it’s a war of attrition and apparently that never occurred to the pentagon command structure. One of the things that Europe learned pretty thoroughly before the Franco-Prussian War, is that an experienced command structure that knows how to communicate and plan is essential. Or so we would think.
I’ll stop soon. Because I doubt the program will live much longer.
I know you’re as fascinated as I am – admit it. It’s like every horrible project we’ve all ever been involved with, rolled up in one, and with the entire budget of several countries at stake. [Read more…]
[WARNING: Long for today’s attention-spans. Readers over 50 should be OK.]
You’ve probably heard that before. Perhaps you’ve heard the same regarding large language models. One thing that this does is casually glosses over the fact that the two approaches work very differently. Or, more precisely, the two approaches are categories of approaches, which can have independent implementation details, as well.
This post almost certainly will not be as well-organized as I wish it could be. I feel that the topic is of great importance, but I am unconfident of my ability to organize an argument, and I am painfully aware that I know far less about the topic than I ought to. Consequently, I welcome dismissive comments as well as substantive ones – part of what I need/want to do is learn more about the topic, and I’m having trouble even figuring out where to start. This will all become clear in a bit, I hope.
Classified materials and their handling appears to still be a news story. Back in 2016, I wrote a bit [stderr] about Clinton (Hillary)’s personal email server, which was accurate as far as I could make it, but that story has been somewhat shaded by recent mis-handling stories. Having to watch journalist and lawyers on the topic can be extremely painful, because of my constant awareness that the information about how classification regimes work is out there and all you need to do is a day or 2 of research.
It’s funny how much effort we put into building redundant and reliable systems (e.g.: “cloud computing”) that scale and replicate well – yet they are subject to the simplest of attacks that can disable them.
… is anyone actually surprised by this? Disappointed, sure – but surprised?