Category: Worth Reading

Another week, another interesting AI article (is anyone writing about anything else these days?), this time from Noah Smith (another author worth following). I found this gem hidden in his weekly roundup:

Instead of trying to write a piece of code from scratch, or prove a math theorem from scratch, or figure out some piece of knowledge for yourself, you just ask AI to do it all for you. So everyone ends up getting the right answers to questions whose answers are already known, so they don’t end up adding anything new.

Phil Gervasi wrote an interesting article describing Rail-Optimized Networking for AI Training Workloads. Go read it first; I’ll wait.

Does it sound interesting? Were you able to see behind the curtain and figure out what it’s really about?

Milan Zapletal, the author of the biggest netlab topology I’ve seen so far (full story), has published a nice introductory article explaining What is Netlab and Why Network Engineers Should Care.

Thank you, Milan! Much appreciated ;)

In a recent article, Scott Alexander made an interesting point: What AI produces are not hallucinations but shameless guesses (also known as bullshit) because the training process rewards the correct answers but does not penalize the incorrect ones. After all, having an AI model say, “I don’t know that” is not good for business, is it?

On a tangential note, calling those blunders hallucinations was a marketing masterstroke. Not being a native English speaker, I might be missing some nuances, but I feel like hallucinations might be something you’re not responsible for (some of the time), whereas we all know who’s responsible for bullshit and shameless guesses – and responsibility is something the AI companies are clearly trying to stay as far away from as possible.

Avery Pennarun published yet another excellent article: every layer of review makes you 10x slower, effectively reiterating what I’ve been saying for decades: all the technology in the world won’t help you unless you re-architect the broken processes.

AI is no exception, but of course, the AI evangelists, LinkedIn AI Wranglers¹, and Thought Leaders will never tell you that (or even admit it).

read more add comment

Geoff Huston published an article supposedly describing the challenge of securing NTP, but as is usually the case, he couldn’t skip the prior art going all the way back (almost) to the formation of Earth.

Before coming to the how do we secure NTP section, you’ll learn everything about the wobbly Earth rotation, the changes in the Earth’s angular speed, the impact of tides, the smearing of leap seconds, the differences between UT1 and UTC, why we use quasars to measure time, and everything there is to know about NTP. Have fun!

The never-ending “we will replace developers” (or networking engineers) pipe dream didn’t start with the latest bout of AI hype (or SDN). As Stephan Schwab explains in his Why We’ve Tried to Replace Developers Every Decade article, it started with COBOL, the magic high-level programming language that businesspeople would use to write their own programs.

At least some of us know how well that ended. I was also unfortunate to be there for the 5GL hype, the forms-driven programming hype, the “everyone will solve every problem out there with Excel macros” (it does work for networking inventory, doesn’t it?), and a few others. So please excuse me if I remain a bit skeptical about the latest fad, even though I find it (like all the previous ones) very useful when used conservatively in limited domains.

Here’s an interesting tidbit from the what took them so long department: Cloudflare One Client continuously measures end-to-end MTU and adjusts the local tunnel interface MTU size accordingly (warning: there’s a fair amount of dubious handwaving over the interesting details), generating ICMP packet-too-big messages as close to the source as possible.

I managed to avoid VPN clients most of my life, so I have no idea whether this is a “finally someone figured that out 🎉” moment or a late catch-up to what other VPN clients have been doing for ages. Feedback (in comments or otherwise) would be most welcome!

Bruce Davie published a nice article explaining why it makes little sense to use an algorithm that’s supposedly faster than Dijkstra’s in link-state routing protocols.

Other interesting data points from the article (and linked presentations):

• People are running (a few) thousands of routers in a single area
• Running Dijkstra’s algorithm on an emulated network with 2000 nodes took 100 msec… in 2003 (page 18 of this NANOG presentation).

It turns out (as I expected) that all the noise about the need for new routing protocols we were experiencing a few years ago was either due to bad implementations or coming from nerds looking for new toys to play with.

Following a link in another Martin Fowler’s blog post, I stumbled upon his thoughts on Open Space events – a way to set up self-organizing events.

I’m not sure I’m brave (or young) enough to try it out, but if you’re planning to organize a small gathering (like a local Network Operator Group), this might be an interesting, slightly more structured approach than a Net::Beer event. It would also be nice to know whether someone managed to pull it off in an online format.