Already have 50 people playing the BSidesPGH CTF! (I obscured the closest screens so don’t try to zoom in for hints 😁)

Such a cool day today at the lockpicking village at BSidesPGH It’s one of my favorite events of the year and helps me recharge and be thankful for this awesome community!

praying for the crowdstrike intern 🙏

My daughter just told me how to plug in a USB-A charger "You plug it in this way, then the other way, then you plug it in the first way you tried and it works. It's so weird" She has discovered the USB-A superposition 🙃

We knew very little about how LLMs actually work...until now. Anthropic just dropped the most insane research paper, detailing some of the ways AI "thinks." And it's completely different than we thought. Here are their wild findings: 🧵

Take your #CloudSecurity knowledge to new heights at the 2025 #CloudSecNextSummit! Connect with industry leaders this October, in person in Denver, CO or Live Online. Secure your seat now: sans.org/u/1AdD youtu.be/dTvh8-6nANo?si…

#BSidesPGH is on Friday July 11th at Rivers. Early bird registration ends April 14th, CFP the 15th. Get registered now!

Come hang on the live stream and talk Driver Exploitation for Offensive Operators with Jake Mayhew youtube.com/live/zK_z4NYVP…

Hold on Kiro still doesn't support OAuth? Or remote MCP servers at all? Come on.

The most interesting part for me is where Andrej Karpathy describes why LLMs aren't able to learn like humans. As you would expect, he comes up with a wonderfully evocative phrase to describe RL: “sucking supervision bits through a straw.” A single end reward gets broadcast across

Looks like Anthropic is giving the Sonnet 5/Opus 4.6 deployment another go. Recovered quite quickly

Go update claude code, it has Opus 4.6 now 🚀

IDE hooks are absolutely 🔥. Up until just a few months ago, only Claude Code & Kiro supported it but now we have Cursor, Windsurf, Auggie, VS Code, and probably more that I'm completely missing It's the number 1 way to get your IDE to do something deterministically, do it 🚀

Is it just me or did Opus 4.6 get massively better at TDD? I used to have to explain it in more detail, it would still write a passing test which tested the broken behavior Now it correctly writes a failing test, fixes code, and then the tests pass and I just say "use TDD"

I'm sorry Claude, I didn't mean it 😬

Is it a legitimate business expense to buy all of the Mad Max movies so I can understand all the gastown references? Asking for a friend

2026, the year where you easily burn through the 50,000 minutes GitHub Enterprise gives you on a project with just two people

Mark your calendars, Jim Manico from Manicode Security and I are gonna be hosting an AI coding nerdery conversation on March 18th See you there 🚀

Holy shit. What a mess, and happens all the time without being transparent enough to blog about it. Kudos Alexey Grigorev Guardrails for AI can help, but even they aren't perfect The most important thing is to do something about it, now. And then improve what you're doing over time.

Code may be free but outages sure aren't Some form of guardrails are more important now than ever