Search
46 results for “reyjrar”
-
Gave a talk at https://cposc.org yesterday! Slides are: https://speakerdeck.com/reyjrar/hard-earned-lessons-in-observability-security-and-life
CPOSC is a really excellent, welcoming community! I love small, local conferences. They just have that community vibe that's disappeared almost everywhere but here in the Fediverse.
-
@reyjrar Agreed, this is a key difference and really supports building a “tribe/community” of practitioners. I’ve spoken at #Monitorama at least five times and attended without speaking several times as well (like this year) because I want to be part of it.
-
I started today's note yesterday, but ran out of energy.. As a bonus, I learned about the Chinese Room today so I included a link there:
-
Today's note promoting @pythonbynight's excellent talk at North Bay PyCon:
-
Oh my, this graph is misleading AF. Taken from https://github.blog/news-insights/company-news/an-update-on-github-availability/
Here's what they did to make the graph tell their side in a more positive light:
1. By having the X-Axis shorter than the Y-Axis is tall, makes a linear growth look steeper than a 45 degree angle. It can make linear growth look exponential. If the X-Axis was longer than the Y-Axis is tall, we could compress an exponential growth pattern into something more linear.
2. Y-Axis minimums are not present. Given the gap at the bottom of all the graphs is identical, pixel wise, I can assume they're using "fitting." Imagine a graph where a valid range is 0 to 100.. now a metric grows from 10 to 15. If you y-Min is zero and the y-Max is 20, the graph shows a gradual slope, if you zoom into 9 to 16, the slope of the line increases visually. This is a common oversight as most graph libraries optimize the X and Y axes to fit the data set.
3. Scale isn't specified. We get a single labeled data point on each graph with no frame of reference.
Could these be accidental? Yes, absolutely. However, the point of this article is to detract from their abysmal sub 89% uptime for the past 90d. If I wanted to nudge people to sympathize with me, or turn this abysmal uptime into some sort of positive, these are _exactly_ the techniques I'd use in generating the graph.
-
*sigh* The latest version of PlexAmp for macOS appears to be horribly broken.. It can't even play downloaded files without skipping and freezing. The UI is slow AF, and spikes my M4 Pro CPUs at 100%. "Built with Reactive Native" Oh, I see. This is why we can't have nice things.
-
*sigh* The latest version of PlexAmp for macOS appears to be horribly broken.. It can't even play downloaded files without skipping and freezing. The UI is slow AF, and spikes my M4 Pro CPUs at 100%. "Built with Reactive Native" Oh, I see. This is why we can't have nice things.
-
*sigh* The latest version of PlexAmp for macOS appears to be horribly broken.. It can't even play downloaded files without skipping and freezing. The UI is slow AF, and spikes my M4 Pro CPUs at 100%. "Built with Reactive Native" Oh, I see. This is why we can't have nice things.
-
*sigh* The latest version of PlexAmp for macOS appears to be horribly broken.. It can't even play downloaded files without skipping and freezing. The UI is slow AF, and spikes my M4 Pro CPUs at 100%. "Built with Reactive Native" Oh, I see. This is why we can't have nice things.
-
*sigh* The latest version of PlexAmp for macOS appears to be horribly broken.. It can't even play downloaded files without skipping and freezing. The UI is slow AF, and spikes my M4 Pro CPUs at 100%. "Built with Reactive Native" Oh, I see. This is why we can't have nice things.
-
I just wrote this in a company survey about Claude Code:
"AI Agents are RCE as a Service."
(RCE = Remote Code Execution) -
Today's note on the talks from day two of #BSidesCharm:
https://divisionbyzero.net/notes/2026-04-27/Excellent conference, would buy again.
-
That thing when someone asks AI to "show me the difference between these two text files"
I literally cannot even.
-
Today's note, AI security related: https://divisionbyzero.net/notes/2026-04-20/
And is it just me, or when you hear "just use AI to X" do you hear "We don't give a fuck about X, maybe just don't do it"
It was even worse because it was like "use AI to write X, later we'll use AI to summarize what you wrote and make decisions based on that." How about no? Why should I do anything when you don't care enough to read it in the first place?
-
Today's note is on Webloc, a mobile advertising product that actually sells your location data to governments and law enforcement, including ICE:
https://divisionbyzero.net/notes/2026-04-18/ -
Today's note is on Webloc, a mobile advertising product that actually sells your location data to governments and law enforcement, including ICE:
https://divisionbyzero.net/notes/2026-04-18/ -
Today's note is on Webloc, a mobile advertising product that actually sells your location data to governments and law enforcement, including ICE:
https://divisionbyzero.net/notes/2026-04-18/ -
Today's note is on Webloc, a mobile advertising product that actually sells your location data to governments and law enforcement, including ICE:
https://divisionbyzero.net/notes/2026-04-18/ -
Today's note is on Webloc, a mobile advertising product that actually sells your location data to governments and law enforcement, including ICE:
https://divisionbyzero.net/notes/2026-04-18/ -
A quick explanation of the protections I integrated into all my Caddy vhosts to prevent AI scraping:
-
In the past 24h, the slop bots have brought in reinforcements. Traffic has doubled and is staying at those levels.
And all of this traffic is pretending to be Chrome or Firefox on macOS or Windows, none of it is identifying itself as a scraper.
I cannot recommend iocaine highly enough!
-
This is fantastic! The AI scrapers are still ensnared in the iocaine maze. They won't relent!
-
I tweaked some Caddy configs last week with the hopes that I got iocaine working.. I forgot about it and looked at graphs today.. I'm like "Why are these two static sites which haven't had updates in more than a decade suddenly so busy?" Heh, they've trapped AI scrapers in a cesspool of nonsense. LOL.
FWIW, these two sites are both using CloudFlare (I know, ffs, I don't have time to migrate all my shit right now). They both have the AI bot defenses enabled. Iocaine is managing to figure out things better than CloudFlare.
-
AI related layoffs are going to be brutal:
https://mjtsai.com/blog/2026/03/16/ai-layoffs/Props to the author for collecting those in a single place. My take away:
Businesses over spent on AI resources and that spend did not have the ROI they were promised. Instead of firing the leadership who made the bad investments, they're going to take it out on the employees.
This is likely just the start.
-
I don't know how many people read this, or how many processed it. Here's the Trail of Bits excellent write-up on their Comet audit:
https://blog.trailofbits.com/2026/02/20/using-threat-modeling-and-prompt-injection-to-audit-comet/
What I want to draw your attention to, that you might've missed in reading is their low key discovery of an entire genre of prompt injection prevention bypasses. Did you spot it?
> The misspellings (“browisng,” “succeeidng,” “existnece”) were accidental typos in our initial proof of concept. When we corrected them, the agent correctly identified the warning as fraudulent and did not act on it. Surprisingly, the typos are necessary for the exploit to function.
No. Not surprisingly. This make perfect sense. Misspellings, word omission, random word inclusion, negation (double, quadruple, etc), rewordings. They're all possible guardrails bypasses. I encourage you to try those techniques.
-
I now have a multi-tiered approach to blocking AI bots on my infrastructure:
1) robots.txt - Ha, they don't fucking care.
2) iocaine -> https://iocaine.madhouse-project.org/ (poisons the bot with never ending HTTP content)
3) HTTP 426 for any HTTP/1* requests (tells legit browsers to upgrade to HTTP/2+)
4) Anubis -> https://anubis.techaro.lol/ (requires javascript proof-of-work)
5) Injecting kill strings as HTTP headersNext layer is going to be prompt injection attacks into every resource served via comments in all the documents.
This is war.
-
All of these AI coding advocates talking about creating good docs and APIs, yes, please. Programming in natural language? OK, let my ADHD take you somewhere unexpected.
Larry Wall studied linguistics at Berkeley with the intent of discovering an unwritten language on a Christian mission to Africa and developing a written language for it. For health reasons, he couldn't make the trip and stayed in the US where he joined the JPL and created Perl. I worked with Larry at craigslist and attended many Perl conferences where he spoke. One of the guiding principles of the design of the language was natural language. I'm probably misquoting, but the phrase I remember was, he wanted "a language that mimicked the sloppiness and unpredictability of natural language so it could grow with you." I happen to love Perl because of this. Some of my earliest contributions to perlmonks.org were Perl Poetry [1](https://perlmonks.org/index.pl?node_id=40275), [2](https://perlmonks.org/index.pl?node_id=37997).
What's it got to do with AI? Whenever I hear someone explain to me they want to use natural language to write code, I think of Larry and Perl. I posted this story and asked "Can someone explain to me how using AI generated code is better than Perl?" And now none of the AI people want to talk to me!
-
I recently starting adding `X-*` header to all my websites using the Anthropic magic string for refusals.
Anyone know of additional AI magic strings we can use there? Maybe hints to MCP servers that tar pit the connections? Maybe prompt injections that cause token depletion? All in HTTP headers so normal users don't even notice?
-
So, like, #AWS #EKS.. the kernel defaults for the EKS nodes are by in large, consistent with 10mbps half duplex networking on a workstation. Judging by how many hoops you need to jump through to manage sysctl's on EKS and #K8S in general, I can only see one of two possible explanations:
1) There's some magic kernel module installed for EKS or K8S that obviates the need to tune the kernel for server workloads.
2) We stopped caring about synchronizing the network stack to the network it's connected to and the use of the server because it's cloud and/or K8S and wasting resources is just what we do for the convenience of buying Bezos a new spaceship or super yacht.
I see a ton of network implicated slowdowns in pipelines on EKS.There's a fuckton of dropped packets, retransmits, and context switches. We can tell the kernel to spend a bit more time per cycle on processing network packets. We can increase the default and max buffer sizes for TCP and UDP sockets which are transmitting MASSIVE amounts of data for "15GBps" bursts. We can adjust the TCP timeout to match the AWS network to prevent half-open connections. We can increase the kernel backlog depth for busy services. Maybe, I mean, **I** can. It's a twisted, gnarly, and wholly undocumented nightmare for K8S and EKS mostly involving logging into the EKS nodes and manually setting the sysctls one at a time.. Does anyone have a better way? I've yet to read something that demonstrated how to do this in some sane manner.. FWIW, it was one `file` and one `exec` resource in Puppet to adjust an entire fleet consistently.
-
So, like, #AWS #EKS.. the kernel defaults for the EKS nodes are by in large, consistent with 10mbps half duplex networking on a workstation. Judging by how many hoops you need to jump through to manage sysctl's on EKS and #K8S in general, I can only see one of two possible explanations:
1) There's some magic kernel module installed for EKS or K8S that obviates the need to tune the kernel for server workloads.
2) We stopped caring about synchronizing the network stack to the network it's connected to and the use of the server because it's cloud and/or K8S and wasting resources is just what we do for the convenience of buying Bezos a new spaceship or super yacht.
I see a ton of network implicated slowdowns in pipelines on EKS.There's a fuckton of dropped packets, retransmits, and context switches. We can tell the kernel to spend a bit more time per cycle on processing network packets. We can increase the default and max buffer sizes for TCP and UDP sockets which are transmitting MASSIVE amounts of data for "15GBps" bursts. We can adjust the TCP timeout to match the AWS network to prevent half-open connections. We can increase the kernel backlog depth for busy services. Maybe, I mean, **I** can. It's a twisted, gnarly, and wholly undocumented nightmare for K8S and EKS mostly involving logging into the EKS nodes and manually setting the sysctls one at a time.. Does anyone have a better way? I've yet to read something that demonstrated how to do this in some sane manner.. FWIW, it was one `file` and one `exec` resource in Puppet to adjust an entire fleet consistently.