I just, take a laptop and a book, and luxuriate in the fact that once I'm at the airport I have no immediate responsibilities to anybody, and it's the one and only time in my life when that's true?

I'm absolutely on team 2-hours-in-advance.

Rip Kirby and a woman walk through the fog.

Look at the pen work on this Alex Raymond RIP KIRBY strip! My gawd.

I'm fine while I'm moving, even doing really strenuous stuff; it's laying still that hurts. That's when all the old injuries remind me that you never really heal, you just compensate.

Strong third parties are absent from our system because of Duverger’s Law. Plurality-winner voting will always make it impossible for third parties to do anything but act as spoilers.

The fact that this just records to an SD card is excellent. I have an Italian mother who loves to tell stories at dinner, but it’s awkward to try to record them. A dumb, easy-to-use and unobtrusive mic is perfect!

Furloughs canceled at Boulder NOAA Global Monitoring Lab after $4 million funding release The reversal follows a weekslong funding lapse that had put 42 researchers at risk of going without pay, threatening work on vital atmospheric and climate science.

“Furlough notices for about half the staff at Boulder’s National Oceanic and Atmospheric Administration Global Monitoring Lab … were rescinded April 16. Work is resuming after more than $4 million in funding was released.”

boulderreportinglab.org/2026/04/16/f...

Also you know there’s a difference between “the capabilities have plateaued” and “they’ve nerfed it”, right?

I’ve seen that sentiment, I just haven’t experienced the same sort of problems.

Internet in Australia is rapidly becoming unusable without a VPN due to age verification. You need to verify your age to use some weather and calculator apps now. Scam apps and phishing sites are starting to realise they can just ask for your ID straight up and people will give it up. Insane shit.

“Director” is the title but it will be a lot of IC work, in addition to being a security point of contact and being in charge of incident response.

You have some infosec background, right? We’re interviewing for a security director right now, @nuttycom.01 on Signal if interested.

We thought people were having a hard time adapting to the rate of change before now, just with the human-constructed internet. Today we’re in a situation where even the extreme neophiles can’t keep up with.

I also don’t want to make it seem like I think this is an unalloyed good or anything. The propaganda uses of AI get more dangerous the better the models get. But we *must* be honest and clear-eyed about their capabilities. bsky.app/profile/bell...

The interaction model with 4.7 has changed somewhat, but it’s capabilities haven’t declined. What makes it feel different is that it’s seemingly more inclined to do more in each autonomous step, which makes it harder to keep up with and harder to keep on the rails.

🤷‍♂️ I’m honestly reporting to you my own experience here. I have been writing software professionally for 27 years now. 6 months ago I had written AI off as not fit for purpose. Today it’s a critical tool. I didn’t change, the models did, and Opus 4.7 is better than 4.6, which was better than 4.5, etc

The wheels are really kind of coming off. warontherocks.com/cogs-of-war/...

Your timing was truly amazing.

“The results are in” isn’t a reason argument when the capabilities are improving with every model release. The results from 6 months ago are utterly irrelevant to the results from the models that came out 3 months ago, which are irrelevant to those from the models that came out this week.

I have a transcript of an AI reasoning out the motivation for certain optimizations in a zero knowledge proving system circuit and producing a proof of their correctness. This is graduate-level stuff that maybe tens or hundreds of people in the world are capable of.

The advancement in mathematics is also shocking. Six months ago these were sloppy plagiarism machines. They aren’t today. bsky.app/profile/grac...

I hear you, but I don’t think you really understand what the actual state of AI capabilities are today. Some of us are snowed under trying to deal with *just how good* AI is at something very similar to reasoning: bsky.app/profile/nutt...

Trailer? Two replacements and a spare?

… I kinds should do that, the tires on my trailer are probably getting a bit rotten.

"Current LLMs are better vulnerability researchers than I am."

The statement from a vulnerability researcher that you should respect:

A slide from the talk that clearly states the thesis: "Let's not be in denial." A lot of us who were LLM skeptics have been forced to reconsider our positions in dramatic fashion.

Important screenshot:

Nicholas Carlini - Black-hat LLMs | [un]prompted 2026 YouTube video by unprompted

They're already having a dramatic effect on those of us who are trying to secure computer systems: www.youtube.com/watch?v=1sd2...

Nicholas Carlini - Black-hat LLMs | [un]prompted 2026 YouTube video by unprompted

If you don't believe that AI is a big deal, I suggest you watch this video of how an LLM discovered a heap overflow in the Linux kernel that's been there since 2003. The AI security apocalypse is here. www.youtube.com/watch?v=1sd2...

I mean, computers have legitimately changed everything.

AI has all but shut down the routes to proving yourself via grunt work in so many industries. All the best juniors I trained came in via that way. They didn’t have the social connections or money to position themselves for the success they proved more than capable of, AI is gutting social mobility

language models are cognitive land and thus should be treated in a georgist way

This is the mathematician equivalent of screaming and running around