I'll be your hanging man. You'll be my queen of swords. That pagan moon shining on the church house door. youtube.com/watch?v=sba0vCzu…

You guys I removed 10k lines of code from projects in the last 8-10 days. Perm underclass for me at -10840 LoC.

"Hey Claude, check to see if the device at this IP address is vulnerable to this CVE." I have bad news guys.

We’ve agreed to a partnership with @SpaceX that will substantially increase our compute capacity. This, along with our other recent compute deals, means that we’ve been able to increase our usage limits for Claude Code and the Claude API.

StackOverflow almost from the beginning suffered from the problem of people answering other peoples' questions like: "Did you do any basic research at all before asking this question? This is a classic Wilbur-Dufus inversion problem any programmer should know. Closing topic." I have found some great helpful answers on this site, but as soon as I could pay money to just ask my apparently stupid questions to Claude or Codex and get reliable answers... I had no reason to interact with people like this anymore.

What did I learn today? DPoP baby. Implementing for QUIC tunnel relays for that real BYOD feel.

This was what iron age Northern Europeans did to cook their meat. Juniper berries, sea salt, caraway, marjoram or thyme, garlic cloves, lingonberries... and more were all different possible ways to season or marinade the meat before burying it.

This is true. I have been learning a great deal about different timer mechanisms for protocols because of Claude. And now I can hand write one in C where I could not before. But you have to know what to ask for. I already kind of knew about them (but not in great detail). So I kept asking more and more questions and had it write examples for me, with expository comments. And then it suggested books I should buy, which I bought and read. Claude didn't hallucinate any part of any of this. It was such a tremendous and satisfying experience!

I'm on the hunt for a new role. Ideally this role is a combined engineering product role. Definitely can be IC. I've come along way in AI tooling, agent building, prototypes that turned into released product features. If you're looking for a GSD high agency and hungry engineer/product combo hit me up.

Is there a reason Opus4.7 is not allowed to configure and troubleshoot routers, switches, and firewalls in advanced network designs, @AnthropicAI ? I'm getting that it's against API usage policy when Opus attempts to ssh into a router.

Continuing my efforts to learn things not just fully off-load, today I'm learning about the tools around Xcode like "Instruments." I have a problem with MacAmp and that is when I slide the volume button the spectrum analyzer freezes until I release the volume slider. So to pinpoint this Claude has helped me learn how 1. Identify how I could troubleshoot this and 2. how to use Instruments. Basically I setup an instrument called "os_signpost" in there I add a debug build and add that build (com.macamp.spike which could be anything you want), set the time I want to record for the instrument I created to capture. Then when I hit record it starts the app. I do my testing (in this case 1. a baseline doing nothing 2. next time sliding the volume slider around) and let it record the events. After I take the events and I have Claude analyze the numbers between each run for me. That way I can see at what layer spikes and look into volume controls and thread starvation. Why? Why not?

If it's data about people, it should be readable by people. If it's code that processes people data, that code should be readable by people.

I need a friend at Broadcom. I need the SDK and programming guides for a long-time EOS/EOL chip. Two chips actually. 15 years old.

This is true. In the beginning, when Iluvitar sang the world into existence, there was a single layer2 device at most hanging off of a layer3 port. What I mean is, that was considered best practice. One hub or switch hanging off of a router port. You didn't attach multiple layer2 devices together. Switching was about bandwidth efficiency, not building layer2 networks.

Also, listen... For 1 second or less interval protocols, if your NOS doesn't implement those as table-based (i.e. "array-based") FSMs, timer wheels, and batched packet processing... then they're doing it wrong. And probably on purpose.

Hi there - Was invited to talk on the CloudGambit/Packet Pushers podcast about the messy reality of AI tooling and expertise. Check it out! Links in the next tweet

Creating a custom kernel for a specific GPU has been a very interesting exercise. I asked claude to help me do it and to track performance issues. I have a bandwidth-starved GPU, but the kernel isn't even using the bandwidth available. To eliminate memory from the equation, we repacked the model so that byte order makes the data immediately consumable upon ingestion into the GPU. Portable formats have to have bytes rearranged in different ways, and that takes GPU cycles. After byte reordering, you need to layout the model as one giant block, padding layers so you get alignment across your memory banks. This gives you maximum memory parallelism. It also prevents some layers being afflicted with a bias that puts most of their weights on a subset (even a single!) bank. With memory out of the way, we found that the kernels themselves aren't coming anywhere near efficient memory bandwidth utilization.