Ian Lewis

Ian Lewis

@ianlewis.org

Engineer from 🇺🇸, PR of 🇯🇵. Personal account. I post about programming and Japan. Self-verified. Prev Google Cloud, slsa.dev, gvisor.dev. Created connpass.com. ian@ianlewis.org

394 followers227 following1.2K posts

It's really annoying when folks create container images that run as root and then drop permissions. It requires that I audit your silly overly complicated s6-overlay based container image to make sure it isn't broken and makes it impossible to use good security contexts in Kubernetes.

I think my philosophy on containers couldn't be farther from what s6 is trying to do. Supervisors inside the container throw out much of the value of using containers because they mask all kinds of failures. A failed s6 service will never show up as failing in the Kubernetes API.

Say you don't know much about Japan without saying you don't know much about Japan.

Quantian@quantian.bsky.social

No. If you go to a sit down sushi bar for dinner in Japan, it is going to run you at least ¥10,000 and probably more like ¥20,000, which is a lot of money for a local. Lunches are cheaper, but it is still absolutely an elite experience normal people do not routinely enjoy.

Folks often say that AI can write code better that them but I think there is an uncanny valley of code that is still better written by humans. AI does well at small, well-directed changes and very large agentic loop harness driven migrations etc. but the middle ground is mostly still slop.

💯

hailey@hailey.at

i think the assumption that LLMs are going to make software more stable and secure are basically for the foreseeable future outright false. LLMs are not producing more secure and stable code unless significant attention is paid to what they ship, which it generally is not. and attacking is ~free.

I keep thinking GITS got a lot of things right about the future. One was that cyberattacks will be performed by AI and the only thing fast enough to fight back is another AI. So we'll probably eventually have traffic storms on the internet where different AI battle it out DDOSing each other.

Laurie Voss@seldo.com

It does in fact seem like the promised wave of LLM-accelerated vulnerability exploits is upon us.

Oh lovely

Russ Cox@swtch.com

The details are now public, as of two hours ago. bugzilla.mozilla.org/show_bug.cgi... The secret P-384 keys for these intermediate CAs are all zeros except the last 8 bits. All not-updated software can be MITM'ed now. IOT boxes that can't or don't update are now trivially MITM'able forever.

Ian Lewis reposted

Writing code hasn’t been the bottleneck in software production since the 1980s; RAD, IDEs, OOP, and off-shoring made code (in corporations) a relatively cheap quasi-commodity long ago. The bottleneck has always been making software that’s actually useful.

Butlerian Jihadist of House Dornick@shimminykricket.blacksky.app

Harvard just dropped a study AI productivity in software engineering and it found the average increase in productivity from using Claude code is.....zero and for exactly the reasons you'd think fion.ac/jellyfish.pdf

I'm trying to set up a simple Rackula instance on my local cluster and for a static web app it sure is a mess. It's a needlessly complicated mess of AI slop. Not to mention the inexplicable bugs you find when you actually try to use it.

I fear that this is the future of AI written applications. It's impossible to debug this thing without looking at the code and trying to figure out why it's unable to handle authentication due to some arcane nginx proxy logic that they needed to include in their Docker containers for some reason.

This kind of thinking is based on the highly flawed assumption that getting something done is always better than nothing. There are much more ways to fuck something up than make it better so the assumption fails in the majority of cases.

Christopher Webb@cwebbonline.com

This is actually hell, right?

The silicon valley ethos of trying many things and learning from failures is built on decades of improvements in software development practice that helped guide engineers towards success and minimize the damage caused by failures.

I think a lot of that culture of maximizing innovation while limiting damage was already eroded by companies choosing to externalize the damage rather than minimizing it but it's a lot worse now with AI because it can rebuild anything from scratch (worse than before) in hours.

The feeling I get when I find a Docker image that rewrites local configuration at run time instead of at build time.

Agents sometimes make up the weirdest metaphors. "That is not an economy. It is a cathedral of bookkeeping with no congregation." I'm sorry, what?

イネス⎔@personhood.removal.surgery

mom can i have peer we have peer at home peer at home: arxiv.org/abs/2609.17320

I upgraded my home lab with a 10in 8U rack. Still more to be done to improve cabling in the back but this will do for now.

One of my NUC machines inexplicably bricked itself yesterday. I didn't do anything all day so it's a complete mystery. Reseated the memory and SSD but that didn't help.

Ian Lewis reposted

Our little one passed away today. He was 18 and half years old and was with us for 14 of those years. The best of dogs. Gutted.

It really sucks that me and my wife will have a small window of time where we could actually travel in the next year or two, but it's going to be stupid expensive because of the orange idiot.

This kind of thing sucks so much. Ads are just so lucrative that they can't just let folks pay to not get them.

IGN@ign.com

Disney+ has altered its user agreement to clarify that ads will be allowed on all subscription tiers. https://bit.ly/4hhOjBo

That reminds me, I saw my first non-media on my tv today too and I had a visceral reaction. An ad for designer clothes on the home screen of the TV...

Since about 2015 I've have ~0 interest in "What's new with Chrome". I think the only feature I use that was added since then is tab groups.

A million *just in the US*.

Chris Hayes@chrislhayes.bsky.social

the experience of COVID really drove this home for me. It wans't the apocalypse, life went on, but a million people died and there were enormous social costs, some of which continue to this day and then when it was done most people were like "let's never think about that again"

antirez@antirez.bsky.social

Witch hunting level: some guy uses AI to reverse engineer an M4 GPU driver for Linux and part of the community that should be for the open source, for the hacking, for the liberation and freedom is against him.

I don't have a Linux desktop environment but I need GnuCash so I run in a VNC server via noVNC behind an IAP proxy in my home lab Kubernetes cluster like a psychopath.

Mac studio will have more than twice the unified memory, but I want to use it as a remotely managed, remote dev server and it's a mac. It's possible but it's more of a pain to maintain than Linux.

I just can't be productive with a time-boxed daily schedule. I need to work without distractions until the work is done. I'll never get anything done if I have to finish by certain time. My brain is just not built that way but I feel myself forced into it more and more as I get older.

I think AI extinction risk is hyperbole, and it's not likely to really cause major bodily harm to people on its own but it could very plausibly be used to develop (bio)weapons with the help of unscrupulous users.

AI swarms hacking other companies is a real potential danger too (self directed or otherwise). I don't see the large models copying themselves in the short term but it's not outside the realm of theoretical possibility. They can also install run-of-the-mill backdoors just like any attacker.

I feel this every time I think about playing RimWorld only to spend the full afternoon curating my mods list.

Matt Keeter@mattkeeter.com

Elden Ring is going great so far: last night, I spent 45 minutes in character creation, then went to bed.

The TLDR newsletters are useful but are so AI driven they often have really blatant errors; articles that don't match the subheadings, emails where the articles don't match the subject etc.